On Effective Classification via Neural Networks

Size: px
Start display at page:

Download "On Effective Classification via Neural Networks"

Transcription

1 On Effective Classification via Neural Networks Bin Cui 1, Anirban Mondal 2, Jialie Shen 3, Gao Cong 4, and Kian-Lee Tan 1 1 Singapore-MIT Alliance, National University of Singapore {cuibin, tankl}@comp.nus.edu.sg 2 University of Tokyo Japan anirban@tkl.iis.u-tokyo.ac.jp 3 University of New South Wales Australia jls@cse.unsw.edu.au 4 The University of Edinburgh UK gao.cong@ed.ac.uk Abstract. For addressing the growing problem of junk on the Internet, this paper proposes an effective classifying and cleansing method in this paper. Incidentally, messages can be modelled as semi-structured documents consisting of a set of fields with pre-defined semantics and a number of variable length free-text fields. Our proposed method deals with both fields having pre-defined semantics as well as variable length free-text fields for obtaining higher accuracy. The main contributions of this work are two-fold. First, we present a new model based on the Neural Network (NN) for classifying personal s. In particular, we treat files as a particular kind of plain text files, the implication being that our feature set is relatively large (since there are thousands of different terms in different files). Second, we propose the use of Principal Component Analysis (PCA) as a preprocessor of NN to reduce the data in terms of both size as well as dimensionality so that the input data become more classifiable and faster for the convergence of the training process used in the NN model. The results of our performance evaluation demonstrate that the proposed algorithm is indeed effective in performing filtering with reasonable accuracy. 1 Introduction The ever-increasing number of Internet users coupled with the widespread proliferation of as one of the fastest and most economical forms of communication have resulted in dramatically increasing number of junk s during the past few years. Consequently, users typically need to spend a non-trivial portion of their valuable time in order to delete junk s. Additionally, junk s can also fill up file server storage space quickly, especially at large sites with thousands of users, who may all be receiving duplicate copies of the same junk mail. To address the problem of growing volumes of junk s, automated methods for filtering are now beginning to be deployed in many commercial products, which typically allow users to define a set of logical rules for filtering junk s. However, these solutions have two serious drawbacks. First, they require the users to be savvy enough to create a set of robust rules for filtering purposes. Second, they require the users to constantly tune and refine the filtering rules in response to the K.V. Andersen, J. Debenham, and R. Wagner (Eds.): DEXA 2005, LNCS 3588, pp , c Springer-Verlag Berlin Heidelberg 2005

2 86 B. Cui et al. changes in junk s over time. Understandably, the problem of filtering junk E- mails is challenging in practice due to the dynamically changing nature of junk and the tremendously large number of s. In essence, an effective filter, which requires minimal manual work from the user, has now become a necessity. The problem of dealing with junk s has been extensively researched. Existing approaches to filtering junk s involve the deployment of data mining techniques [6,7], the usage of addresses [9] and the application of text classification techniques [5,1]. In the realm of text classification, an message is viewed as a document and a judgement of its ness is viewed as a class label associated with the document. While text classification has beenextensivelyresearched [4,3,12], empirical study on the document type of and the features of building an effective personal filter within the framework of text classification has received relatively little attention. In this regard, the main contributions of this paper are two-fold: 1. We present a new model based on the Neural Network (NN) for classifying personal s. In particular, we treat files as a specific kind of plain text files, the implication being that our feature set is relatively large (since there are thousands of different terms in different files). 2. We propose the use of Principle Component Analysis (PCA) as a preprocessor of NN to reduce the data in terms of both size and dimensionality so that the input data becomes more classifiable. This facilitates fast convergence of the training process used in the NN model. Notably, PCA only pre-processes the input to NN. The results of our extensive performance evaluation on a real dataset of personal s demonstrate that our proposed method is indeed effective in providing reasonable performance in terms of recall, precision and total accuracy rate, especially for s. To our knowledge, this is the first work on classification that uses a neural network-based strategy for classifying s. The remainder of the paper is organized as follows. Section 2 discusses existing related works, while Section 3 provides the details of our design for the NN method to classify s. Section 4 reports the results of our performance evaluation. Finally, we conclude in Section 5 with directions for future work. 2 Related Work This section reviews related works in the area of junk filtering [2,5,13], Neural Networks [8] and Principle Component Analysis [10]. The Bayesian approach to filtering junk [13] considered domain specific features as well as the raw text of messages and enhanced the performance of a Bayesian classifier by hand-crafting and incorporating many features that are indicative of junk . Representing each individual message as a binary vector, the proposal in [13] detects junk mail in a straightforward manner using a given pre-classified set of training messages. In [2], the authors compare methods for learning text classifiers focussing on the kinds of classification problems that might arise in filtering personal messages. In [5], the documents to be classified are regarded as semistructured textual documents comprising two parts. While one part is a set of structured

3 On Effective Classification via Neural Networks 87 fields with well-defined semantics, the other is a number of variable-length sections of free text. However, not many text classifiers take both portions into consideration. Additionally, conventional classification techniques may not be effective when dealing with variable-length free text. Notably, our work differs from the proposals in [2,5,13] in that these works focus on language processing, while we focus on general electronic text classification. Neural networks [8,11] have been widely adopted in various fields of applications such as pattern recognition and identification. A neural network consists of simple elements i.e., the neurons operating in parallel and the connections between them. The training of a neural network is to adjust the weights of the connections for minimizing the difference between the output of the neural network and the target of the training data. We adopt the supervised back-propagation neural network in our classification system. The advantage of a neural network arises from its computing power (due to its massively parallel distributed structure), its ability to learn and more importantly, its capability to generalize. We only need to design the network structure and then input the training data. The results may be affected by the selection of the network structure and the input attributes, the training data and the stopping criteria. PCA [10] is a widely used method for applications in signal/image filtering and pattern classification. It can transform data in the original space into another feature space, reduce the dimensionality of the input data, while keeping the most significant information. It examines the variance structure in the dataset and determines the directions along which the data exhibits high variance. The first principal component accounts for as much of the variability in the data as possible, and each succeeding component accounts for as much of the remaining variability as possible. Working as a pre-processor of neural networks, it can make the input data more classifiable and reduce the dimensionality of the training and validation dataset by using only the first several features, thereby speeding up the convergence of the training process. 3 System Design This section discusses the design of our proposed system. 3.1 Model of NN for Classification This section discusses how the NN method is used for filtering. The processing via the NN method involves three steps, namely data pre-processing, training and testing, as depicted in Figure 1. The FeatureExtractionrefers to the data pre-processing, the details concerning which will be discussed shortly. For the training dataset, once we obtain the selected features, we feed them into the neural network and generate an E- mail classifier. For each test , we use the classifier to verify the efficiency of NN model. We adopt the error Back-propagation training algorithm in our model since it is one of the simplest as well as one of the most useful neural network algorithms. Now let us discuss the method used for presenting data to the network for training as well as for testing. We employ cross-validation and early stopping. The available data is divided into the following three disjoint sets:

4 88 B. Cui et al. Training: data Feature Feature vectors NN Learning Testing: data Extraction Feature vectors Classifier Output Fig. 1. The overview of NN method 1. Training set: This data is used to train the network. 2. Validation set: The error of the network averaged over this data is used to decide when the training algorithm has found the best approximation to the data without overfitting. 3. Testing set: The best network, determined by the validation test, is applied to the test set and the performance over the test set is given. Training is accomplished by calculating the derivative of the network s error with respect to each weight in the network when presented a particular input pattern. This derivative indicates which direction each weight should be adjusted to reduce the error. Each weight is modified by taking a small step in this direction. With a nonzero momentum factor, a fraction of the previous weight change is added to the new weight value. Notably, this accelerates learning in some cases. The patterns in the training set are traversed one-by-one. A pass through all the training patterns is called an epoch. The training data is repetitively presented for multiple epochs, until a specified number of epochs have been reached. After each epoch, the error of the network applied to the validation set of patterns is calculated. If the current network scores the lowest error so far on the validation set, this network s weights are saved. At the conclusion of training, the network s best weights are used to calculate the network s error on the testing set. Data Pre-processing. In general, messages are semi-structured documents that possess a set of structured fields with predefined semantics and a number of variablelength free text fields. The headers of s are structured fields and usually contain information pertaining to the document, such as sender, date, domain etc. The major contents of the are variable-length free text fields, such as subject and the body. Understandably, both the structured fields and the free text portion could contain important information which could help in determining the class to which an belongs. Therefore, an effective classifier should be able to include features from both the structured fields and the free text. To make optimal use of the information in an message, we generate two kinds of features for each which we use in our NN model. The details are as follows. Structured features: Features represented by structured fields in the header part of an document. In our model, seven structured features were generated.

5 On Effective Classification via Neural Networks 89 Attachment: Attachment occurs in the , true; else false. Content type: If the content type of is plain text, true; else false. Sender Domain: We extract the sender domain from header, if it contains edu, 1; contains com 2; else 3. FW: Subject of header starts with word FW, true; else false. Re: Subject of header starts with word Re, true; else false. To group: The is sent to a group, not a single person, true; else false. CC. The content of CC in the header of s is not empty, true; else false. Textual features: We use general text processing method for dealing with the textual features. The terms occurring in the body of a given and in the Subject of the are extracted and pre-processed, and are regarded as the features of the body of s. We use Document Frequency Threshold to remove those features that have little influence on classification work. Finally, the feature values of the terms are the term weights calculated by the simple idf.tf method. We also use Document Frequency Threshold to remove some features that have little influence on classification work. The threshold should not be too low since we should try to find fewer important features for NN according to the characteristic of NN itself. Training of NN Model. With the results of data pre-processing, we can do the training and generate a classifier to filter the un After processing both the header and body of all the training samples, a 2- D array of feature vector is obtained for and un s. The feature vector includes the features extracted from both the header as well as the body of the s. The class labels of the s are also recorded in a 2- D array. Class label is 1 for s and 0 for un s. 2. Use the Error Back-propagation algorithm for generating a classifier from the feature vectors and then the classifier is returned as a file. Testing of NN Model. In this stage, we will test the efficiency of our classifier. 1. Generate the feature vector for the header of each , group the feature vector of header and body together for each , and form a complete feature vector for each . The process is same as training part. 2. According to the feature vector generated for each new , use classifier, which has been built in training stage, to compute the output score for each . If the score > threshold (specified by the users), this can be regarded as the , or else un Principal Component Analysis Finding principal components (PCs) is basically a mathematical problem of finding the principal singular vectors of the input dataset using the well-known singular value decomposition method. The PCA transformation has two steps. First, we do principal component analysis of the training and validation data set. In this step, we transform the training and validation dataset into the singular vector space and get the eigenvectors

6 B. Cui et al. and eigenvalues of each dimension. Second, we transform the testing data set into the same space as the training and validation data. This is done by simply multiplying it with the eigenvector matrix produced in the first step. In order to have some idea of the data after the PCA transformation, we plot the first two principal vectors of a training dataset with 747 features after transformation in Figure 2. The points are the vectors of the s, while the circles are the vectors of the un E- mails. From the figure, we observe that by using only the first two features, the vectors corresponding to the and un s can be distinguished relatively easily. Figure 3 depicts the eigenvalues of the principal components. It is clear that the first few eigenvalues are much more important than the later ones Eigenvalue No. of dimensions Fig. 2. Plotting of the first two PCsof the 747-features dataset Fig. 3. Plotting of the eigenvalues of the 747- features dataset 4 Performance Evaluation This section reports the performance evaluation of our proposed NN model for classification. All the experiments have been conducted on a SUN E450 machine with SUN OS 5.7. We have used a total of 2000 real personal s as the dataset for our experiments. Notably, a dataset of 2K s is considered large in our context primarily because of the difficulty associated with the collection of personal s. We manually label each as or un i.e., s are divided into two classes for our experiments. For the dataset used in our experiments, the number of s and the number of un s were 1500 and 500 respectively. During the experiments, the whole dataset was divided into three portions randomly, namely training data, validation data and testing data. While the training data is used to train a classifier, the validation data scores the error of training and provides the best model, which is then classified by the testing data. We have used precision and recall as the metrics for evaluating the performance of our proposed classification approach because all the methods provide very fast response. Although the training stage of NN is time-consuming, the filtering stage of NN is extremely efficient and typically requires less than one second, hence we do not present response time as a metric in this paper. The definitions of recall and precision

7 On Effective Classification via Neural Networks 91 for and un s are quite similar. Here, we provide the definition for s only: recall = Nii Nii N, precision = N i ;wheren is the number of total s, N i is the number of s classified as,andn ii is the number of s classified as. The output of neural network is a float from 0 to 1, and we classify an as only when the output is greater than some pre-defined threshold. We label the 1 and un 0, hence when we test the , the output of approaches 1 and the output of un approaches 0 if we make a correct classification. Initially, we set the threshold as a default value 0.5. In all the experiments, we set a large epoch number in the training stage and only use the best weights in testing phase. 4.1 Effect of Neuron Number First, we test the effect of neuron number of our neural network model on the classification performance. Basically, the neural network model has two hidden layers, and we can tune the number of neurons in the both of hidden layers of the network. The neuron number in the first layer must be larger than 1; while the neuron number of the second layer can be 0, in this case the neural network has only one layer. Denote neuron pair as n1/n2,wheren1 is the number of first layer neuron and n2 is the number of second layer neuron. precision(%) recall(%) un 1/0 1/1 1/5 5/0 5/1 5/5 10/0 10/1 10/5 No. of neurons (a) Precision un 1/0 1/1 1/5 5/0 5/1 5/5 10/0 10/1 10/5 No. of neurons (b) Recall Fig. 4. Effect of neuron number on different layers The first experiment aims at the performance of the classifier when different neuron pairs are used. The performance of NN of different neuron number is shown in Figure sets of features and 9 neuron pairs are used in our experiment to test which pair of neurons can perform well with high precision and high recall. For each neuron pair, Figure 4 (a) and (b) show the average precision and average recall of all these feature sets, respectively. A few observations can be made from this experiment. First, the precision and recall curves of s are better than the ones of un s no matter which neuron pair is used. Second, it is obvious that neuron pairs 1/5 and 10/0 can perform better than other neuron pairs for both the s and

8 92 B. Cui et al. un s. The average recall for both and un s is above %. Finally, it is easier for neural network to identify files in our experiment. The reason could be that we select more files than un ones. Hence they can be clearly characterized by some features. In the following experiments, we will fix the structure of the NN model according to the performance shown in Figure 4. The NN has two hidden layers: the first layer has one neuron and the second layer has 5 neurons. 4.2 Effect of Feature Selection As mentioned earlier, we extracted two kinds of features from s, structured features and textual features. 8 structured features and up to 00 textual features are generated for files. One question under exploration is how important these features are in classification. From the experiment, we can see how many features are better and enough to classify s and how these features influence the recall, precision and accuracy rate of classification. The performance of NN for different feature selection is shown in Figure 5. precision(%) recall(%) un No. of features (a) Precision un No. of features (b) Recall Fig. 5. Effect of feature selection 10 feature sets are selected to study the performance of NN classifier. The sizes of 10 feature sets are 8, 28, 37, 50, 92, 121, 243, 508, 747, and 1313 respectively. Each feature set consists of all 8 structured features and some textual features selected by tf.idf formula. For example, 28 stands for 8 structured features plus 20 textual features. Figure 5 presents the precision and recall of different features for and un s. The results show that the performanceof classification is very unstable when the number of features we use is fewer than 200. Especially, when we only use structured features, the recall of un s is surprisingly high, %; on the contrary, the recall of s is very low, 79%. However, the overall recall and precision for classification increase and become more stable when we use more features. With the number of features increasing, the accuracy rate increase gradually. From the figures, we can see that although 1313 features provide the best performance, the 508 features perform only a little worse and it can save more than

9 On Effective Classification via Neural Networks 93 half space to save the features and weights, and the execution time is less than half. So based on the figures, we select 508 features as an optimal strategy. The results of the experiment show that the feature selection plays an important role in classification. We will use the PCA method to do the feature selection below and see how PCA influences the performance of classification more clearly. 4.3 Effect of PCA The Performance of NN after PCA processing is shown in Figure 6. We use 1313 features of dataset, and do the principal component analysis and generate the new feature space with different dimensions (PCs). precision(%) recall(%) un No. of dimensions (a) Precision Fig. 6. Effect of PCA un No. of dimensions (b) Recall Using PCA, most of the information in the original space is condensed into a few dimensions along which the variances in the data distribution are the largest. PCA method aims to transform the feature space into a new space there the first dimensions are more important, so we can easily select the more important features to do the classification effectively and efficiently. From the results, we can see that the average precision and recall rate is more than 93%, compared with about 85% when the PCA method is not used, for 8 features. Additionally, the number is low, say less than 200, the performance after PCA processing is more stable and better. Although in the second experiment, we select the most important words for each feature selection, the PCA can capture more information. When the number is 37, the performance is optimal and the space and time cost is only 10% of NN without PCA. Furthermore, we also find that fewer features selected by PCA can describe the characteristic of s as well as the whole original features. It shows that adding many unimportant features does not necessarily enhance the performance of classification. 4.4 Comparison with other Schemes We also compared our NN model with the decision tree and Naive Bayesian Classifier methods used in [5,13]. Because of different feature selections, we only compare the optimal performance for three methods. To simplify the comparison, we adopt the error number of false classification rate, which is number of classfied messages. The error rate of proposed NN model is only 4%, which is more than 50% better than two competitive methods.

10 94 B. Cui et al. 5 Conclusion In examining the growing problem of dealing with junk , we provide NN model which embeds PCA as a preprocessor to eliminate junk from a user s mail stream. The efficiency of such filters can also be greatly enhanced by considering not only the full text of the messages, but also a set of structural features. Different ways of feature selection for the model were evaluated. Performance of the classifier was compared with respect to feature selection, parameter tuning and PCA processing. Our experiment results show that our models provide good performance in filtering junk s. In the near future, we plan to incorporate other techniques (e.g., address analysis, real-time user feedback) into our proposed NN model. Moreover, we intend to examine the cost-effective integration of our classification scheme into existing systems. References 1. X. Carreras and L. Marquez. Boosting trees for anti-spam filtering. In Proc. Recent Advances in Natural Language Processing, W. W. Cohen. Learning rules that classify . Proc. the AAAI Spring Symposium on Machine Learning in Information Access, W.W. Cohen and Y. Singer. Context-sensitive learning methods for text categorization. Proc. SIGIR, M. Craven, D. DiPasquo, D. Freitag, A. McCallum, T. Mitchell, K. Nigam, and S. Slattery. Learning to extract symbolic knowledge from the world wide web. Proc. the 15th National Conference on Artificial Intelligence, Y. L. Diao, H. J. Lu, and D. K. Wu. A comparative study of classification based personal filtering. Proc. the 4th PAKDD, T. Fawcett. in vivo spam filtering: A challenge problem for data mining. In KDD Explorations vol.5 no.2, K. R. Gee. Using latent semantic indexing to filter spam. In ACM Symposium on Applied Computing, Data Mining Track, S. Haykin. Neural networks: A comprehensive foundation. International Ed., Prentice-Hall, 2nd Ed, J. Ioannidis. Fighting spam by encapsulating policy in addresses. In Proc. Network and Distributed Systems Security Conference (NDSS), I. T. Jolliffe. Principle Componet Analysis. Springer-Verlag, S. Y. Kung. Digital neural networks. Prentice-Hall, D. D. Lewis and M. Ringuette. A comparison of two learning algorithms for text categorization. Third Annual Symposium on Document Analysis and Information Retrieval,, M. Sahami, S. Dumais, D. Heckerman, and E. Horvitz. A bayesian approach to filtering junk . Proc. AAAI Workshop Learning for Text Categorization, 1998.

An Empirical Performance Comparison of Machine Learning Methods for Spam Categorization

An Empirical Performance Comparison of Machine Learning Methods for Spam  Categorization An Empirical Performance Comparison of Machine Learning Methods for Spam E-mail Categorization Chih-Chin Lai a Ming-Chi Tsai b a Dept. of Computer Science and Information Engineering National University

More information

A Comparative Study of Classification Based Personal Filtering

A Comparative Study of Classification Based Personal  Filtering A Comparative Study of Classification Based Personal E-mail Filtering Yanlei Diao, Hongjun Lu and Deai Wu Department of Computer Science The Hong Kong University of Science and Technology Clear Water Bay,

More information

A Comparison of Text-Categorization Methods applied to N-Gram Frequency Statistics

A Comparison of Text-Categorization Methods applied to N-Gram Frequency Statistics A Comparison of Text-Categorization Methods applied to N-Gram Frequency Statistics Helmut Berger and Dieter Merkl 2 Faculty of Information Technology, University of Technology, Sydney, NSW, Australia hberger@it.uts.edu.au

More information

LRLW-LSI: An Improved Latent Semantic Indexing (LSI) Text Classifier

LRLW-LSI: An Improved Latent Semantic Indexing (LSI) Text Classifier LRLW-LSI: An Improved Latent Semantic Indexing (LSI) Text Classifier Wang Ding, Songnian Yu, Shanqing Yu, Wei Wei, and Qianfeng Wang School of Computer Engineering and Science, Shanghai University, 200072

More information

The Comparative Study of Machine Learning Algorithms in Text Data Classification*

The Comparative Study of Machine Learning Algorithms in Text Data Classification* The Comparative Study of Machine Learning Algorithms in Text Data Classification* Wang Xin School of Science, Beijing Information Science and Technology University Beijing, China Abstract Classification

More information

Contents Machine Learning concepts 4 Learning Algorithm 4 Predictive Model (Model) 4 Model, Classification 4 Model, Regression 4 Representation

Contents Machine Learning concepts 4 Learning Algorithm 4 Predictive Model (Model) 4 Model, Classification 4 Model, Regression 4 Representation Contents Machine Learning concepts 4 Learning Algorithm 4 Predictive Model (Model) 4 Model, Classification 4 Model, Regression 4 Representation Learning 4 Supervised Learning 4 Unsupervised Learning 4

More information

A novel supervised learning algorithm and its use for Spam Detection in Social Bookmarking Systems

A novel supervised learning algorithm and its use for Spam Detection in Social Bookmarking Systems A novel supervised learning algorithm and its use for Spam Detection in Social Bookmarking Systems Anestis Gkanogiannis and Theodore Kalamboukis Department of Informatics Athens University of Economics

More information

Robustness of Selective Desensitization Perceptron Against Irrelevant and Partially Relevant Features in Pattern Classification

Robustness of Selective Desensitization Perceptron Against Irrelevant and Partially Relevant Features in Pattern Classification Robustness of Selective Desensitization Perceptron Against Irrelevant and Partially Relevant Features in Pattern Classification Tomohiro Tanno, Kazumasa Horie, Jun Izawa, and Masahiko Morita University

More information

Keywords : Bayesian, classification, tokens, text, probability, keywords. GJCST-C Classification: E.5

Keywords : Bayesian,  classification, tokens, text, probability, keywords. GJCST-C Classification: E.5 Global Journal of Computer Science and Technology Software & Data Engineering Volume 12 Issue 13 Version 1.0 Year 2012 Type: Double Blind Peer Reviewed International Research Journal Publisher: Global

More information

Accuracy Analysis of Neural Networks in removal of unsolicited s

Accuracy Analysis of Neural Networks in removal of unsolicited  s Accuracy Analysis of Neural Networks in removal of unsolicited e-mails P.Mohan Kumar P.Kumaresan S.Yokesh Babu Assistant Professor (Senior) Assistant Professor Assistant Professor (Senior) SITE SITE SCSE

More information

AM 221: Advanced Optimization Spring 2016

AM 221: Advanced Optimization Spring 2016 AM 221: Advanced Optimization Spring 2016 Prof. Yaron Singer Lecture 2 Wednesday, January 27th 1 Overview In our previous lecture we discussed several applications of optimization, introduced basic terminology,

More information

COMBINING NEURAL NETWORKS FOR SKIN DETECTION

COMBINING NEURAL NETWORKS FOR SKIN DETECTION COMBINING NEURAL NETWORKS FOR SKIN DETECTION Chelsia Amy Doukim 1, Jamal Ahmad Dargham 1, Ali Chekima 1 and Sigeru Omatu 2 1 School of Engineering and Information Technology, Universiti Malaysia Sabah,

More information

A ew Algorithm for Community Identification in Linked Data

A ew Algorithm for Community Identification in Linked Data A ew Algorithm for Community Identification in Linked Data Nacim Fateh Chikhi, Bernard Rothenburger, Nathalie Aussenac-Gilles Institut de Recherche en Informatique de Toulouse 118, route de Narbonne 31062

More information

Fig 1. Overview of IE-based text mining framework

Fig 1. Overview of IE-based text mining framework DiscoTEX: A framework of Combining IE and KDD for Text Mining Ritesh Kumar Research Scholar, Singhania University, Pacheri Beri, Rajsthan riteshchandel@gmail.com Abstract: Text mining based on the integration

More information

Center for Automation and Autonomous Complex Systems. Computer Science Department, Tulane University. New Orleans, LA June 5, 1991.

Center for Automation and Autonomous Complex Systems. Computer Science Department, Tulane University. New Orleans, LA June 5, 1991. Two-phase Backpropagation George M. Georgiou Cris Koutsougeras Center for Automation and Autonomous Complex Systems Computer Science Department, Tulane University New Orleans, LA 70118 June 5, 1991 Abstract

More information

On-line Estimation of Power System Security Limits

On-line Estimation of Power System Security Limits Bulk Power System Dynamics and Control V, August 26-31, 2001, Onomichi, Japan On-line Estimation of Power System Security Limits K. Tomsovic A. Sittithumwat J. Tong School of Electrical Engineering and

More information

Image Classification Using Wavelet Coefficients in Low-pass Bands

Image Classification Using Wavelet Coefficients in Low-pass Bands Proceedings of International Joint Conference on Neural Networks, Orlando, Florida, USA, August -7, 007 Image Classification Using Wavelet Coefficients in Low-pass Bands Weibao Zou, Member, IEEE, and Yan

More information

Improving Recognition through Object Sub-categorization

Improving Recognition through Object Sub-categorization Improving Recognition through Object Sub-categorization Al Mansur and Yoshinori Kuno Graduate School of Science and Engineering, Saitama University, 255 Shimo-Okubo, Sakura-ku, Saitama-shi, Saitama 338-8570,

More information

A Semi-Supervised Approach for Web Spam Detection using Combinatorial Feature-Fusion

A Semi-Supervised Approach for Web Spam Detection using Combinatorial Feature-Fusion A Semi-Supervised Approach for Web Spam Detection using Combinatorial Feature-Fusion Ye Tian, Gary M. Weiss, Qiang Ma Department of Computer and Information Science Fordham University 441 East Fordham

More information

Using Query History to Prune Query Results

Using Query History to Prune Query Results Using Query History to Prune Query Results Daniel Waegel Ursinus College Department of Computer Science dawaegel@gmail.com April Kontostathis Ursinus College Department of Computer Science akontostathis@ursinus.edu

More information

Using Real-valued Meta Classifiers to Integrate and Contextualize Binding Site Predictions

Using Real-valued Meta Classifiers to Integrate and Contextualize Binding Site Predictions Using Real-valued Meta Classifiers to Integrate and Contextualize Binding Site Predictions Offer Sharabi, Yi Sun, Mark Robinson, Rod Adams, Rene te Boekhorst, Alistair G. Rust, Neil Davey University of

More information

ECE 285 Class Project Report

ECE 285 Class Project Report ECE 285 Class Project Report Based on Source localization in an ocean waveguide using supervised machine learning Yiwen Gong ( yig122@eng.ucsd.edu), Yu Chai( yuc385@eng.ucsd.edu ), Yifeng Bu( ybu@eng.ucsd.edu

More information

Detecting Spam Bots in Online Social Networking Sites: A Machine Learning Approach

Detecting Spam Bots in Online Social Networking Sites: A Machine Learning Approach Detecting Spam Bots in Online Social Networking Sites: A Machine Learning Approach Alex Hai Wang College of Information Sciences and Technology, The Pennsylvania State University, Dunmore, PA 18512, USA

More information

Identifying Important Communications

Identifying Important Communications Identifying Important Communications Aaron Jaffey ajaffey@stanford.edu Akifumi Kobashi akobashi@stanford.edu Abstract As we move towards a society increasingly dependent on electronic communication, our

More information

A Network Intrusion Detection System Architecture Based on Snort and. Computational Intelligence

A Network Intrusion Detection System Architecture Based on Snort and. Computational Intelligence 2nd International Conference on Electronics, Network and Computer Engineering (ICENCE 206) A Network Intrusion Detection System Architecture Based on Snort and Computational Intelligence Tao Liu, a, Da

More information

arxiv: v1 [cs.lg] 12 Feb 2018

arxiv: v1 [cs.lg] 12 Feb 2018 Email Classification into Relevant Category Using Neural Networks arxiv:1802.03971v1 [cs.lg] 12 Feb 2018 Deepak Kumar Gupta & Shruti Goyal Co-Founders: Reckon Analytics deepak@reckonanalytics.com & shruti@reckonanalytics.com

More information

Keyword Extraction by KNN considering Similarity among Features

Keyword Extraction by KNN considering Similarity among Features 64 Int'l Conf. on Advances in Big Data Analytics ABDA'15 Keyword Extraction by KNN considering Similarity among Features Taeho Jo Department of Computer and Information Engineering, Inha University, Incheon,

More information

CLASSIFICATION WITH RADIAL BASIS AND PROBABILISTIC NEURAL NETWORKS

CLASSIFICATION WITH RADIAL BASIS AND PROBABILISTIC NEURAL NETWORKS CLASSIFICATION WITH RADIAL BASIS AND PROBABILISTIC NEURAL NETWORKS CHAPTER 4 CLASSIFICATION WITH RADIAL BASIS AND PROBABILISTIC NEURAL NETWORKS 4.1 Introduction Optical character recognition is one of

More information

A modified and fast Perceptron learning rule and its use for Tag Recommendations in Social Bookmarking Systems

A modified and fast Perceptron learning rule and its use for Tag Recommendations in Social Bookmarking Systems A modified and fast Perceptron learning rule and its use for Tag Recommendations in Social Bookmarking Systems Anestis Gkanogiannis and Theodore Kalamboukis Department of Informatics Athens University

More information

Radial Basis Function Neural Network Classifier

Radial Basis Function Neural Network Classifier Recognition of Unconstrained Handwritten Numerals by a Radial Basis Function Neural Network Classifier Hwang, Young-Sup and Bang, Sung-Yang Department of Computer Science & Engineering Pohang University

More information

A Rough Set Approach for Generation and Validation of Rules for Missing Attribute Values of a Data Set

A Rough Set Approach for Generation and Validation of Rules for Missing Attribute Values of a Data Set A Rough Set Approach for Generation and Validation of Rules for Missing Attribute Values of a Data Set Renu Vashist School of Computer Science and Engineering Shri Mata Vaishno Devi University, Katra,

More information

Data Compression. The Encoder and PCA

Data Compression. The Encoder and PCA Data Compression The Encoder and PCA Neural network techniques have been shown useful in the area of data compression. In general, data compression can be lossless compression or lossy compression. In

More information

An Intelligent Clustering Algorithm for High Dimensional and Highly Overlapped Photo-Thermal Infrared Imaging Data

An Intelligent Clustering Algorithm for High Dimensional and Highly Overlapped Photo-Thermal Infrared Imaging Data An Intelligent Clustering Algorithm for High Dimensional and Highly Overlapped Photo-Thermal Infrared Imaging Data Nian Zhang and Lara Thompson Department of Electrical and Computer Engineering, University

More information

A *69>H>N6 #DJGC6A DG C<>C::G>C<,8>:C8:H /DA 'D 2:6G, ()-"&"3 -"(' ( +-" " " % '.+ % ' -0(+$,

A *69>H>N6 #DJGC6A DG C<>C::G>C<,8>:C8:H /DA 'D 2:6G, ()-&3 -(' ( +-   % '.+ % ' -0(+$, The structure is a very important aspect in neural network design, it is not only impossible to determine an optimal structure for a given problem, it is even impossible to prove that a given structure

More information

Rank Measures for Ordering

Rank Measures for Ordering Rank Measures for Ordering Jin Huang and Charles X. Ling Department of Computer Science The University of Western Ontario London, Ontario, Canada N6A 5B7 email: fjhuang33, clingg@csd.uwo.ca Abstract. Many

More information

Query-Sensitive Similarity Measure for Content-Based Image Retrieval

Query-Sensitive Similarity Measure for Content-Based Image Retrieval Query-Sensitive Similarity Measure for Content-Based Image Retrieval Zhi-Hua Zhou Hong-Bin Dai National Laboratory for Novel Software Technology Nanjing University, Nanjing 2193, China {zhouzh, daihb}@lamda.nju.edu.cn

More information

An Efficient Approach for Color Pattern Matching Using Image Mining

An Efficient Approach for Color Pattern Matching Using Image Mining An Efficient Approach for Color Pattern Matching Using Image Mining * Manjot Kaur Navjot Kaur Master of Technology in Computer Science & Engineering, Sri Guru Granth Sahib World University, Fatehgarh Sahib,

More information

Performance Improvement of Hardware-Based Packet Classification Algorithm

Performance Improvement of Hardware-Based Packet Classification Algorithm Performance Improvement of Hardware-Based Packet Classification Algorithm Yaw-Chung Chen 1, Pi-Chung Wang 2, Chun-Liang Lee 2, and Chia-Tai Chan 2 1 Department of Computer Science and Information Engineering,

More information

An Algorithm For Training Multilayer Perceptron (MLP) For Image Reconstruction Using Neural Network Without Overfitting.

An Algorithm For Training Multilayer Perceptron (MLP) For Image Reconstruction Using Neural Network Without Overfitting. An Algorithm For Training Multilayer Perceptron (MLP) For Image Reconstruction Using Neural Network Without Overfitting. Mohammad Mahmudul Alam Mia, Shovasis Kumar Biswas, Monalisa Chowdhury Urmi, Abubakar

More information

Similarity Image Retrieval System Using Hierarchical Classification

Similarity Image Retrieval System Using Hierarchical Classification Similarity Image Retrieval System Using Hierarchical Classification Experimental System on Mobile Internet with Cellular Phone Masahiro Tada 1, Toshikazu Kato 1, and Isao Shinohara 2 1 Department of Industrial

More information

Fast Fuzzy Clustering of Infrared Images. 2. brfcm

Fast Fuzzy Clustering of Infrared Images. 2. brfcm Fast Fuzzy Clustering of Infrared Images Steven Eschrich, Jingwei Ke, Lawrence O. Hall and Dmitry B. Goldgof Department of Computer Science and Engineering, ENB 118 University of South Florida 4202 E.

More information

Visual object classification by sparse convolutional neural networks

Visual object classification by sparse convolutional neural networks Visual object classification by sparse convolutional neural networks Alexander Gepperth 1 1- Ruhr-Universität Bochum - Institute for Neural Dynamics Universitätsstraße 150, 44801 Bochum - Germany Abstract.

More information

Efficient SQL-Querying Method for Data Mining in Large Data Bases

Efficient SQL-Querying Method for Data Mining in Large Data Bases Efficient SQL-Querying Method for Data Mining in Large Data Bases Nguyen Hung Son Institute of Mathematics Warsaw University Banacha 2, 02095, Warsaw, Poland Abstract Data mining can be understood as a

More information

Singular Value Decomposition, and Application to Recommender Systems

Singular Value Decomposition, and Application to Recommender Systems Singular Value Decomposition, and Application to Recommender Systems CSE 6363 Machine Learning Vassilis Athitsos Computer Science and Engineering Department University of Texas at Arlington 1 Recommendation

More information

Optimized Implementation of Logic Functions

Optimized Implementation of Logic Functions June 25, 22 9:7 vra235_ch4 Sheet number Page number 49 black chapter 4 Optimized Implementation of Logic Functions 4. Nc3xe4, Nb8 d7 49 June 25, 22 9:7 vra235_ch4 Sheet number 2 Page number 5 black 5 CHAPTER

More information

Using PageRank in Feature Selection

Using PageRank in Feature Selection Using PageRank in Feature Selection Dino Ienco, Rosa Meo, and Marco Botta Dipartimento di Informatica, Università di Torino, Italy {ienco,meo,botta}@di.unito.it Abstract. Feature selection is an important

More information

Graph Matching Iris Image Blocks with Local Binary Pattern

Graph Matching Iris Image Blocks with Local Binary Pattern Graph Matching Iris Image Blocs with Local Binary Pattern Zhenan Sun, Tieniu Tan, and Xianchao Qiu Center for Biometrics and Security Research, National Laboratory of Pattern Recognition, Institute of

More information

Bayesian Spam Detection System Using Hybrid Feature Selection Method

Bayesian Spam Detection System Using Hybrid Feature Selection Method 2016 International Conference on Manufacturing Science and Information Engineering (ICMSIE 2016) ISBN: 978-1-60595-325-0 Bayesian Spam Detection System Using Hybrid Feature Selection Method JUNYING CHEN,

More information

Applying Supervised Learning

Applying Supervised Learning Applying Supervised Learning When to Consider Supervised Learning A supervised learning algorithm takes a known set of input data (the training set) and known responses to the data (output), and trains

More information

Training Digital Circuits with Hamming Clustering

Training Digital Circuits with Hamming Clustering IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS I: FUNDAMENTAL THEORY AND APPLICATIONS, VOL. 47, NO. 4, APRIL 2000 513 Training Digital Circuits with Hamming Clustering Marco Muselli, Member, IEEE, and Diego

More information

A Content Vector Model for Text Classification

A Content Vector Model for Text Classification A Content Vector Model for Text Classification Eric Jiang Abstract As a popular rank-reduced vector space approach, Latent Semantic Indexing (LSI) has been used in information retrieval and other applications.

More information

Detecting Printed and Handwritten Partial Copies of Line Drawings Embedded in Complex Backgrounds

Detecting Printed and Handwritten Partial Copies of Line Drawings Embedded in Complex Backgrounds 9 1th International Conference on Document Analysis and Recognition Detecting Printed and Handwritten Partial Copies of Line Drawings Embedded in Complex Backgrounds Weihan Sun, Koichi Kise Graduate School

More information

FACE RECOGNITION USING SUPPORT VECTOR MACHINES

FACE RECOGNITION USING SUPPORT VECTOR MACHINES FACE RECOGNITION USING SUPPORT VECTOR MACHINES Ashwin Swaminathan ashwins@umd.edu ENEE633: Statistical and Neural Pattern Recognition Instructor : Prof. Rama Chellappa Project 2, Part (b) 1. INTRODUCTION

More information

Face Recognition using Eigenfaces SMAI Course Project

Face Recognition using Eigenfaces SMAI Course Project Face Recognition using Eigenfaces SMAI Course Project Satarupa Guha IIIT Hyderabad 201307566 satarupa.guha@research.iiit.ac.in Ayushi Dalmia IIIT Hyderabad 201307565 ayushi.dalmia@research.iiit.ac.in Abstract

More information

SELECTION OF THE OPTIMAL PARAMETER VALUE FOR THE LOCALLY LINEAR EMBEDDING ALGORITHM. Olga Kouropteva, Oleg Okun and Matti Pietikäinen

SELECTION OF THE OPTIMAL PARAMETER VALUE FOR THE LOCALLY LINEAR EMBEDDING ALGORITHM. Olga Kouropteva, Oleg Okun and Matti Pietikäinen SELECTION OF THE OPTIMAL PARAMETER VALUE FOR THE LOCALLY LINEAR EMBEDDING ALGORITHM Olga Kouropteva, Oleg Okun and Matti Pietikäinen Machine Vision Group, Infotech Oulu and Department of Electrical and

More information

Character Recognition Using Convolutional Neural Networks

Character Recognition Using Convolutional Neural Networks Character Recognition Using Convolutional Neural Networks David Bouchain Seminar Statistical Learning Theory University of Ulm, Germany Institute for Neural Information Processing Winter 2006/2007 Abstract

More information

Performance Analysis of Data Mining Classification Techniques

Performance Analysis of Data Mining Classification Techniques Performance Analysis of Data Mining Classification Techniques Tejas Mehta 1, Dr. Dhaval Kathiriya 2 Ph.D. Student, School of Computer Science, Dr. Babasaheb Ambedkar Open University, Gujarat, India 1 Principal

More information

Influence of Neighbor Size for Initial Node Exchange of SOM Learning

Influence of Neighbor Size for Initial Node Exchange of SOM Learning FR-E3-3 SCIS&ISIS2006 @ Tokyo, Japan (September 20-24, 2006) Influence of Neighbor Size for Initial Node Exchange of SOM Learning MIYOSHI Tsutomu Department of Information and Knowledge Engineering, Tottori

More information

A Data Classification Algorithm of Internet of Things Based on Neural Network

A Data Classification Algorithm of Internet of Things Based on Neural Network A Data Classification Algorithm of Internet of Things Based on Neural Network https://doi.org/10.3991/ijoe.v13i09.7587 Zhenjun Li Hunan Radio and TV University, Hunan, China 278060389@qq.com Abstract To

More information

Image retrieval based on bag of images

Image retrieval based on bag of images University of Wollongong Research Online Faculty of Informatics - Papers (Archive) Faculty of Engineering and Information Sciences 2009 Image retrieval based on bag of images Jun Zhang University of Wollongong

More information

Supervised classification of law area in the legal domain

Supervised classification of law area in the legal domain AFSTUDEERPROJECT BSC KI Supervised classification of law area in the legal domain Author: Mees FRÖBERG (10559949) Supervisors: Evangelos KANOULAS Tjerk DE GREEF June 24, 2016 Abstract Search algorithms

More information

6. NEURAL NETWORK BASED PATH PLANNING ALGORITHM 6.1 INTRODUCTION

6. NEURAL NETWORK BASED PATH PLANNING ALGORITHM 6.1 INTRODUCTION 6 NEURAL NETWORK BASED PATH PLANNING ALGORITHM 61 INTRODUCTION In previous chapters path planning algorithms such as trigonometry based path planning algorithm and direction based path planning algorithm

More information

Noise-based Feature Perturbation as a Selection Method for Microarray Data

Noise-based Feature Perturbation as a Selection Method for Microarray Data Noise-based Feature Perturbation as a Selection Method for Microarray Data Li Chen 1, Dmitry B. Goldgof 1, Lawrence O. Hall 1, and Steven A. Eschrich 2 1 Department of Computer Science and Engineering

More information

CSC 2515 Introduction to Machine Learning Assignment 2

CSC 2515 Introduction to Machine Learning Assignment 2 CSC 2515 Introduction to Machine Learning Assignment 2 Zhongtian Qiu(1002274530) Problem 1 See attached scan files for question 1. 2. Neural Network 2.1 Examine the statistics and plots of training error

More information

Content Based Image Retrieval system with a combination of Rough Set and Support Vector Machine

Content Based Image Retrieval system with a combination of Rough Set and Support Vector Machine Shahabi Lotfabadi, M., Shiratuddin, M.F. and Wong, K.W. (2013) Content Based Image Retrieval system with a combination of rough set and support vector machine. In: 9th Annual International Joint Conferences

More information

Enhancing Forecasting Performance of Naïve-Bayes Classifiers with Discretization Techniques

Enhancing Forecasting Performance of Naïve-Bayes Classifiers with Discretization Techniques 24 Enhancing Forecasting Performance of Naïve-Bayes Classifiers with Discretization Techniques Enhancing Forecasting Performance of Naïve-Bayes Classifiers with Discretization Techniques Ruxandra PETRE

More information

International Journal of Scientific Research & Engineering Trends Volume 4, Issue 6, Nov-Dec-2018, ISSN (Online): X

International Journal of Scientific Research & Engineering Trends Volume 4, Issue 6, Nov-Dec-2018, ISSN (Online): X Analysis about Classification Techniques on Categorical Data in Data Mining Assistant Professor P. Meena Department of Computer Science Adhiyaman Arts and Science College for Women Uthangarai, Krishnagiri,

More information

STUDYING OF CLASSIFYING CHINESE SMS MESSAGES

STUDYING OF CLASSIFYING CHINESE SMS MESSAGES STUDYING OF CLASSIFYING CHINESE SMS MESSAGES BASED ON BAYESIAN CLASSIFICATION 1 LI FENG, 2 LI JIGANG 1,2 Computer Science Department, DongHua University, Shanghai, China E-mail: 1 Lifeng@dhu.edu.cn, 2

More information

Classification of Face Images for Gender, Age, Facial Expression, and Identity 1

Classification of Face Images for Gender, Age, Facial Expression, and Identity 1 Proc. Int. Conf. on Artificial Neural Networks (ICANN 05), Warsaw, LNCS 3696, vol. I, pp. 569-574, Springer Verlag 2005 Classification of Face Images for Gender, Age, Facial Expression, and Identity 1

More information

Learning and Generalization in Single Layer Perceptrons

Learning and Generalization in Single Layer Perceptrons Learning and Generalization in Single Layer Perceptrons Neural Computation : Lecture 4 John A. Bullinaria, 2015 1. What Can Perceptrons do? 2. Decision Boundaries The Two Dimensional Case 3. Decision Boundaries

More information

COLLABORATIVE AGENT LEARNING USING HYBRID NEUROCOMPUTING

COLLABORATIVE AGENT LEARNING USING HYBRID NEUROCOMPUTING COLLABORATIVE AGENT LEARNING USING HYBRID NEUROCOMPUTING Saulat Farooque and Lakhmi Jain School of Electrical and Information Engineering, University of South Australia, Adelaide, Australia saulat.farooque@tenix.com,

More information

A Modular k-nearest Neighbor Classification Method for Massively Parallel Text Categorization

A Modular k-nearest Neighbor Classification Method for Massively Parallel Text Categorization A Modular k-nearest Neighbor Classification Method for Massively Parallel Text Categorization Hai Zhao and Bao-Liang Lu Department of Computer Science and Engineering, Shanghai Jiao Tong University, 1954

More information

SSV Criterion Based Discretization for Naive Bayes Classifiers

SSV Criterion Based Discretization for Naive Bayes Classifiers SSV Criterion Based Discretization for Naive Bayes Classifiers Krzysztof Grąbczewski kgrabcze@phys.uni.torun.pl Department of Informatics, Nicolaus Copernicus University, ul. Grudziądzka 5, 87-100 Toruń,

More information

Managing Open Bug Repositories through Bug Report Prioritization Using SVMs

Managing Open Bug Repositories through Bug Report Prioritization Using SVMs Managing Open Bug Repositories through Bug Report Prioritization Using SVMs Jaweria Kanwal Quaid-i-Azam University, Islamabad kjaweria09@yahoo.com Onaiza Maqbool Quaid-i-Azam University, Islamabad onaiza@qau.edu.pk

More information

ADAPTIVE LINK WEIGHT ASSIGNMENT AND RANDOM EARLY BLOCKING ALGORITHM FOR DYNAMIC ROUTING IN WDM NETWORKS

ADAPTIVE LINK WEIGHT ASSIGNMENT AND RANDOM EARLY BLOCKING ALGORITHM FOR DYNAMIC ROUTING IN WDM NETWORKS ADAPTIVE LINK WEIGHT ASSIGNMENT AND RANDOM EARLY BLOCKING ALGORITHM FOR DYNAMIC ROUTING IN WDM NETWORKS Ching-Lung Chang, Yan-Ying, Lee, and Steven S. W. Lee* Department of Electronic Engineering, National

More information

11/14/2010 Intelligent Systems and Soft Computing 1

11/14/2010 Intelligent Systems and Soft Computing 1 Lecture 7 Artificial neural networks: Supervised learning Introduction, or how the brain works The neuron as a simple computing element The perceptron Multilayer neural networks Accelerated learning in

More information

Identifying Layout Classes for Mathematical Symbols Using Layout Context

Identifying Layout Classes for Mathematical Symbols Using Layout Context Rochester Institute of Technology RIT Scholar Works Articles 2009 Identifying Layout Classes for Mathematical Symbols Using Layout Context Ling Ouyang Rochester Institute of Technology Richard Zanibbi

More information

Combination of PCA with SMOTE Resampling to Boost the Prediction Rate in Lung Cancer Dataset

Combination of PCA with SMOTE Resampling to Boost the Prediction Rate in Lung Cancer Dataset International Journal of Computer Applications (0975 8887) Combination of PCA with SMOTE Resampling to Boost the Prediction Rate in Lung Cancer Dataset Mehdi Naseriparsa Islamic Azad University Tehran

More information

Slides for Data Mining by I. H. Witten and E. Frank

Slides for Data Mining by I. H. Witten and E. Frank Slides for Data Mining by I. H. Witten and E. Frank 7 Engineering the input and output Attribute selection Scheme-independent, scheme-specific Attribute discretization Unsupervised, supervised, error-

More information

Review on Methods of Selecting Number of Hidden Nodes in Artificial Neural Network

Review on Methods of Selecting Number of Hidden Nodes in Artificial Neural Network Available Online at www.ijcsmc.com International Journal of Computer Science and Mobile Computing A Monthly Journal of Computer Science and Information Technology IJCSMC, Vol. 3, Issue. 11, November 2014,

More information

Texture Classification by Combining Local Binary Pattern Features and a Self-Organizing Map

Texture Classification by Combining Local Binary Pattern Features and a Self-Organizing Map Texture Classification by Combining Local Binary Pattern Features and a Self-Organizing Map Markus Turtinen, Topi Mäenpää, and Matti Pietikäinen Machine Vision Group, P.O.Box 4500, FIN-90014 University

More information

Unsupervised Learning

Unsupervised Learning Networks for Pattern Recognition, 2014 Networks for Single Linkage K-Means Soft DBSCAN PCA Networks for Kohonen Maps Linear Vector Quantization Networks for Problems/Approaches in Machine Learning Supervised

More information

Application of Support Vector Machine Algorithm in Spam Filtering

Application of Support Vector Machine Algorithm in  Spam Filtering Application of Support Vector Machine Algorithm in E-Mail Spam Filtering Julia Bluszcz, Daria Fitisova, Alexander Hamann, Alexey Trifonov, Advisor: Patrick Jähnichen Abstract The problem of spam classification

More information

Combining Review Text Content and Reviewer-Item Rating Matrix to Predict Review Rating

Combining Review Text Content and Reviewer-Item Rating Matrix to Predict Review Rating Combining Review Text Content and Reviewer-Item Rating Matrix to Predict Review Rating Dipak J Kakade, Nilesh P Sable Department of Computer Engineering, JSPM S Imperial College of Engg. And Research,

More information

Face Detection Using Radial Basis Function Neural Networks with Fixed Spread Value

Face Detection Using Radial Basis Function Neural Networks with Fixed Spread Value IJCSES International Journal of Computer Sciences and Engineering Systems, Vol., No. 3, July 2011 CSES International 2011 ISSN 0973-06 Face Detection Using Radial Basis Function Neural Networks with Fixed

More information

A Neural Network Model Of Insurance Customer Ratings

A Neural Network Model Of Insurance Customer Ratings A Neural Network Model Of Insurance Customer Ratings Jan Jantzen 1 Abstract Given a set of data on customers the engineering problem in this study is to model the data and classify customers

More information

Automatic Domain Partitioning for Multi-Domain Learning

Automatic Domain Partitioning for Multi-Domain Learning Automatic Domain Partitioning for Multi-Domain Learning Di Wang diwang@cs.cmu.edu Chenyan Xiong cx@cs.cmu.edu William Yang Wang ww@cmu.edu Abstract Multi-Domain learning (MDL) assumes that the domain labels

More information

Improved Classification of Known and Unknown Network Traffic Flows using Semi-Supervised Machine Learning

Improved Classification of Known and Unknown Network Traffic Flows using Semi-Supervised Machine Learning Improved Classification of Known and Unknown Network Traffic Flows using Semi-Supervised Machine Learning Timothy Glennan, Christopher Leckie, Sarah M. Erfani Department of Computing and Information Systems,

More information

1. Introduction. 2. Motivation and Problem Definition. Volume 8 Issue 2, February Susmita Mohapatra

1. Introduction. 2. Motivation and Problem Definition. Volume 8 Issue 2, February Susmita Mohapatra Pattern Recall Analysis of the Hopfield Neural Network with a Genetic Algorithm Susmita Mohapatra Department of Computer Science, Utkal University, India Abstract: This paper is focused on the implementation

More information

Creating Situational Awareness with Spacecraft Data Trending and Monitoring

Creating Situational Awareness with Spacecraft Data Trending and Monitoring Creating Situational Awareness with Spacecraft Data Trending and Monitoring Zhenping Li ASRC Technical Services, Zhenping.Li@asrcfederal.com J.P. Douglas, and Ken Mitchell ASRC Technical Services, JPaul.Douglas@noaa.gov

More information

1 Machine Learning System Design

1 Machine Learning System Design Machine Learning System Design Prioritizing what to work on: Spam classification example Say you want to build a spam classifier Spam messages often have misspelled words We ll have a labeled training

More information

Some questions of consensus building using co-association

Some questions of consensus building using co-association Some questions of consensus building using co-association VITALIY TAYANOV Polish-Japanese High School of Computer Technics Aleja Legionow, 4190, Bytom POLAND vtayanov@yahoo.com Abstract: In this paper

More information

NEURAL NETWORK-BASED SEGMENTATION OF TEXTURES USING GABOR FEATURES

NEURAL NETWORK-BASED SEGMENTATION OF TEXTURES USING GABOR FEATURES NEURAL NETWORK-BASED SEGMENTATION OF TEXTURES USING GABOR FEATURES A. G. Ramakrishnan, S. Kumar Raja, and H. V. Raghu Ram Dept. of Electrical Engg., Indian Institute of Science Bangalore - 560 012, India

More information

FACE RECOGNITION USING FUZZY NEURAL NETWORK

FACE RECOGNITION USING FUZZY NEURAL NETWORK FACE RECOGNITION USING FUZZY NEURAL NETWORK TADI.CHANDRASEKHAR Research Scholar, Dept. of ECE, GITAM University, Vishakapatnam, AndraPradesh Assoc. Prof., Dept. of. ECE, GIET Engineering College, Vishakapatnam,

More information

Towards Rule Learning Approaches to Instance-based Ontology Matching

Towards Rule Learning Approaches to Instance-based Ontology Matching Towards Rule Learning Approaches to Instance-based Ontology Matching Frederik Janssen 1, Faraz Fallahi 2 Jan Noessner 3, and Heiko Paulheim 1 1 Knowledge Engineering Group, TU Darmstadt, Hochschulstrasse

More information

Lecture 2 Notes. Outline. Neural Networks. The Big Idea. Architecture. Instructors: Parth Shah, Riju Pahwa

Lecture 2 Notes. Outline. Neural Networks. The Big Idea. Architecture. Instructors: Parth Shah, Riju Pahwa Instructors: Parth Shah, Riju Pahwa Lecture 2 Notes Outline 1. Neural Networks The Big Idea Architecture SGD and Backpropagation 2. Convolutional Neural Networks Intuition Architecture 3. Recurrent Neural

More information

Evolving SQL Queries for Data Mining

Evolving SQL Queries for Data Mining Evolving SQL Queries for Data Mining Majid Salim and Xin Yao School of Computer Science, The University of Birmingham Edgbaston, Birmingham B15 2TT, UK {msc30mms,x.yao}@cs.bham.ac.uk Abstract. This paper

More information

Assignment 2. Classification and Regression using Linear Networks, Multilayer Perceptron Networks, and Radial Basis Functions

Assignment 2. Classification and Regression using Linear Networks, Multilayer Perceptron Networks, and Radial Basis Functions ENEE 739Q: STATISTICAL AND NEURAL PATTERN RECOGNITION Spring 2002 Assignment 2 Classification and Regression using Linear Networks, Multilayer Perceptron Networks, and Radial Basis Functions Aravind Sundaresan

More information

Deep Tensor: Eliciting New Insights from Graph Data that Express Relationships between People and Things

Deep Tensor: Eliciting New Insights from Graph Data that Express Relationships between People and Things Deep Tensor: Eliciting New Insights from Graph Data that Express Relationships between People and Things Koji Maruhashi An important problem in information and communications technology (ICT) is classifying

More information

4.12 Generalization. In back-propagation learning, as many training examples as possible are typically used.

4.12 Generalization. In back-propagation learning, as many training examples as possible are typically used. 1 4.12 Generalization In back-propagation learning, as many training examples as possible are typically used. It is hoped that the network so designed generalizes well. A network generalizes well when

More information