A proposal of hierarchization method based on data distribution for multi class classification

Size: px

Start display at page:

Download "A proposal of hierarchization method based on data distribution for multi class classification"

Neil Hensley
6 years ago
Views:

1 1 2 2 Paul Horton 3 2 One-Against-One OAO One-Against-All OAA UCI k-nn OAO SVM A proposal of hierarchization method based on data distribution for multi class classification Tatsuya Kusuda, 1 Shinya Watanabe, 2 Jianming Shi 2 and Paul Horton 3 This paper proposes a new hierarchical method for multi-class classification using binary classifiers. Unlike existing extension methods, such as One- Against-One (OAO) and One-Against-All (OAA), our proposed method makes a hierarchic structure of classification according to distributions of each class data as given training data. Thus, a more accurate multi-class classification can be expected than by existing methods. In particular, our proposed method is expected to be more effective for classes with few samples. Since a hierarchical structure of classification derived by our proposed method is formed on the basis of similarities among classes, relative features of each class can be inferable through the classification structure. In this paper, the effectiveness of our proposed method is discussed through some examples from UCI repository, based on comparison with that of k-nn and OAO. 1. SVM 1) 4) k k-nearest neighbor algorithm k-nn SVM 2 2 (Support Vector Machine, SVM) 1) 2 One-Against-One OAO One-Against-All OAA 5) 6) OAO OAA 5) 1 Graduate School of Information and Electronic Engineering, Muroran Institute of Technology 2 Department of Information and Electronic Engineering, Muroran Institute of Technology 3 National Institute of Advanced Industrial Science and Technology Computational Biology Research Center (AIST CBRC) 1

2 UCI 7) k-nn SVM OAO 2 3 UCI hierarchy hierarchy hierarchy classifier 3 1 classifier 2 3 classifier4 class 2 Conceptual diagram of proposal method classifier class 2 3 classifier5 class Step1: Step2: 2.2 Step3: Step2 SVM Step4: Step3 Step2 Step1:. Step2: SVM Step3: OAO SVM Step means k-means k=2 2

3 2-means 2-means 8) ,1, Step1: 2-means 2 Step2: Step1 Step3: α% Step4: N Gj Step5: C G Balance Accuracy Eval = β Balance + (1 β) Accuracy (1) Balance = G j=1 ( N Gj G C ) (2) Accuracy = Step6: Step1 Step5 (3) Step7: Step1 Step6 Step8: (1) 1 2 β 0.3 Step3 α α 3. UCI k-nn SVM OAO 3.1 UCI 7) 6 Iris,Wine,Heart Disease,Glass,Vowel,Car Evaluation α β Balance (1) 2-means 2 Car Evaluation α 0.2 α UCI 6 Iris Wine Heart Disease Glass Vowel Car Evaluation 1 Car Evaluation

4 1 The characteristics of the used data Dataset Number Features Classes of data Iris Wine Heart Disease Glass Vowel Car Evaluation Parameter 2 Used parameters Determination of group parameter α Values Weight of evaluation 0.3 formula parameter β Number of partitions 10 in cross validation Number of clustering 30 4 Glass The predictive accuracy of each class in Glass Id Number of data Weighted k-nn OAO Proposed method % 71.43% 84.29% % 76.32% 71.05% % 11.76% 17.65% % % 46.15% % 22.22% 44.44% % 79.31% 86.21% % 66.36% 70.56% 3 The results of predictive accuracy Dataset Weighted k-nn OAO Proposed method Iris 95.33% 96.00% 97.33% Wine 93.26% 97.19% 97.75% Heart Disease 55.56% 55.22% 56.57% Glass 65.42% 66.36% 70.56% Vowel 92.02% 99.60% 95.04% Car Evaluation 88.60% 99.36% 99.07% (10-fold Cross-Validation) k-nn SVM OAO SVM Iris Wine Heart Disease Glass Vowel Car Evaluation OAO Vowel Car Evaluation OAO α Glass Glass 4 4 Glass 2 OAO 5 α The effects of α value for predictive accuracy Dataset Weighted OAO Proposed method k-nn α values Iris 94.00% 94.67% 96.00% 95.33% 95.33% 96.00% 96.67% Wine 93.26% 97.19% 97.75% 97.19% 97.75% 98.31% 97.19% Heart Disease 54.88% 54.88% 54.88% 54.88% 55.22% 54.88% 54.88% Glass 65.42% 64.95% 64.49% % 64.95% 67.76% 64.95% Vowel 92.02% 99.60% 95.45% 95.15% 95.05% 95.75% 99.60% Car Evaluation 82.81% 99.71% 99.42% 99.54% 99.71% 99.71% 99.71% α α α α α 5 5 α α OAO 4

5 id 6 Glass Classes of Glass data Details of classes 1 building windows float processed 2 building windows non float processed 3 vehicle windows float processed 5 containers 6 tableware 7 headlamps 7 Glass The attributes of Glass data id Features name Details of features 1 RI refractive index 2 Na Sodium 3 Mg Magnesium 4 Al Aluminum 5 Si Silicon 6 K Potassium 7 Ca Calcium 8 Ba Barium 9 Fe Iron OAO Glass Heart Disease Glass UCI 6 7 Glass Heart Disease Heart Disease id 9 2 Glass Hierarchy diagram of proposal technique(glass) Classes id Heart Disease Classes of Heart Disease data Details of classes 0 person who hasn t heart disease person who has heart disease 1 4 is degree 9 Heart Disease The attributes of Heart Disease data Features id Features name Details of features 1 age age in years 2 sex sex (1 = male; 0 = female) 3 cp chest pain type 4 trestbps resting blood pressure 5 chol serum cholestoral in mg/dl 6 fbs fasting blood sugar 120 mg/dl 7 restecg resting electrocardiographic results 8 thalach maximum heart rate achieved 9 exang exercise induced angina 10 oldpeak ST depression induced by exercise relative to rest 11 slope the slope of the peak exercise ST segment 12 ca number of major vessels (0-3) colored by flourosopy 13 thal 3 = normal; 6 = fixed defect; 7 = reversable defect 5

6 3 Heart Disease 0.2 Hierarchy diagram of proposal technique(heart Disease 0.2) 4 (Heart Disease 0.4) Hierarchy diagram of proposal technique(heart Disease 0.4) α UCI k-nn OAO OAO 1) G. Nalbantov P.J.F.Groenen and J.C.Bioch. a majorization approach to linear support vector machines with different hinge errors. Advances in Data Analysis and Classification, Vol. 2, pp , ) J.A.K. Suykens and J. Vandewalle: Least squares support vector machine classifiers, Neural Netherlands, Vol. 9, No. 3, pp , ) Doumpos, M. Zopounidis, C. Golfinopoulou, V: Additive Support Vector Machines for Pattern Classification, IEEE Transactions on Systems, Man, and Cybernetics, Part B: Cybernetics, Vol. 37, No. 3, pp , ) Fukumizu, K: Special statistical properties of neural network learning. Proc, NOLTA 97, pp , ) Jonathan Milgram and Mohamed Cheriet and Robert Sabourin: One Against One or One Against All :Which One is Better for Handwriting Recognition with SVMs?, Ecole de Technologie Superieure, Montreal, Canada, ) Chih-Wei Hsu Chih-Jen Lin: A comparison of methods for multiclass support vector machines, Neural Networks, IEEE Transactions on, Vol. 13, No. 2, pp , ) C.L.Blake and C.J.Merz: UCI repository of machine learning databases, University of California, Department of Information and Computer Science, 1998, 8),,,,1988 9),, ) Baldi P et al: Assessing the accuracy of prediction algorithms for classification, an overview, Bioinformatics, Vol. 16, No. 5, pp ,

Predicting Diabetes and Heart Disease Using Diagnostic Measurements and Supervised Learning Classification Models

Predicting Diabetes and Heart Disease Using Diagnostic Measurements and Supervised Learning Classification Models Kunal Sharma CS 4641 Machine Learning Abstract Supervised learning classification algorithms