Time Series Data Analysis on Agriculture Food Production

Transcription

1 , pp Time Series Data Analysis on Agriculture Food Production A.V.S. Pavan Kumar 1 and R. Bhramaramba 2 1 Research Scholar, Department of Computer Science and Engineering, GIT, GITAM University, Visakhapatnam avspavankumarmca@gmail.com 2 Associate Professor, Department of Information Technology, GIT, GITAM University, Visakhapatnam bhramarambaravi@gmail.com Abstract. Agriculture food production is the backbone of every economy. In a country like India, which has ever increasing demand of food due to rising population, advances in agriculture sector are required to meet the needs. Time series data is a special case of time stamped data. It is similar to a number line. The events are uniformly separated in time variety of domains like engineering, research, medicine and finance. This paper presents an identifying the underlying structure of the time series and fitting an appropriate Auto Regressive Integrated Moving Average model with a case study on agriculture food production time series data with R software. Keywords: Time Series Data, Agriculture food production data, Prediction, ARIMA Model, Augmented Dickey-Fuller Test, Kwiatkowski-Phillips- Schmidt-Shin test. 1 Introduction Agriculture food production is a large portion of the economic output. Together with the breeding industry, researchers try to identify the food production growth in a yearly manner. The application of machine learning to agriculture is an incredibly important advance in agriculture technologies because it allows these systems to leverage and combine historic information with real-time information such as agriculture crop yield prediction, food production, and weather information, that the farmer maintains. Machine learning techniques augment this heterogeneous data by combining it with the farmer s knowledge to provide coherence to the vast amounts of data being generated in today s farm environment. It allows these systems to learn about characteristics of each field and adapt systemic recommendations that get better over time. Machine learning allows farmers and agribusinesses alike to make better decisions, in real-time, even in the absence of complete information. Time Series data is a special case of time stamped data. It is similar to a number line. The events are uniformly separated in time in variety of domains like engineering, research, medicine and finance. Time series analysis attempts to model the underlying structure of observations taken over time. A time series, denoted Y=a +bx, is an ordered sequence ISSN: ASTL Copyright 2017 SERSC

2 of equally spaced values over time. Time series analysis has several applications in finance, economics, engineering and manufacturing. The important steps involved in time series analysis are a) Identify and model the structure of the time series b) Estimating the Parameters c) Diagnostic Checking with Tests and d) Forecast the future values in the time series. 2 Implementation The ARIMA model is also called as Box-Jenkins methodology (Box and Jenkins 1976). The Box-Jenkins procedure is concerned with fitting a mixed ARIMA model to a given set of data. The main objective in fitting ARIMA model is to identify the stochastic process of the time series and predict the future values accurately. This paper discusses the implementation of forecasting model on food production data with R Tool. The main requirements to implement this are packages named Forecast, fma, expsmooth, lmtest, zoo, tseries need to be loaded into the R studio. The yearly production of food grains in CSV form is the input file. The important steps involved are. 2.1 Examining and converting the continuous data into time series data The data can be understood by using descriptive statistical functions like mean, median, min, max, standard deviation, and summary data. Data in the dataset is in the normal numerical form (Continuous form). So, now data has to be converted into time series data. ts() function in R can be used for the conversion of data from continuous to time series and stored as an object. Once you have read the time series data into R, then store the time series object in R as foodgraints. > foodgraints=ts(foodgrains$food.grain.production,start=1951) To understand the time series data effectively we can use a graph. To draw a graph we use a command called plot function in R studio > plot(foodgraints). Copyright 2017 SERSC 521

3 Fig. 1. Continuous Data Sets to Time Lines 2.2 Model fitting for the analysis of the data ARIMA model needs the series data must be stationary. Stationary data consists of its mean, variance, and auto covariance which are time invariant. The ADF stands for Augmented Dickey-Fuller. This test is a formal stationary test for the data set. > adf.test(foodgraints) KPSS another unit root test called Kwiatkowski-Phillips-Schmidt-Shin test. > kpss.test(foodgraints) # Second Test for Stationary We are creating an ARIMA model. ARIMA stands for Auto-regressive Integrated Moving Average. To create a First ARIMA model, We have a special function called ARIMA function in R studio. This function will fit the data set into arima and creates a model. >fit=arima(foodgraints,order=c(0,1,1)) This created model called fit is added as an object in the Global Environment of the R studio. ACF stands for Auto Correlation Function. This is used to find whether the data is stationary or not. To draw a acf plot use a function called acf() in R studio. > acf(diff(foodgraints)) plot(forecast(fit)) # PACF plot of residuals 522 Copyright 2017 SERSC

4 Fig. 2. Forecast of food grains using ARIMA > Box.test(residuals(fit),type="Ljung") # Ljung Box Test Box-Ljung test data: residuals(fit) X-squared = , df = 1, p-value = Creating new models using ARIMA > fit1=auto.arima(diff(foodgraints)) # Auto Arima for finding the Arima(p,d,q), p,d,qprameters It creates a new model called fit1 and adds to the Global Environment window of the R studio. > plot(forecast(fit1)) # Forecast plot Copyright 2017 SERSC 523

5 Fig. 3. Forecast from ARIMA with non-zero mean > fit2=nnetar(diff(foodgraints),maxit=1000) # Implementing Neural Network on Timeseries data It creates a new object fit2 and adds to the Global Environment window of a R studio. Forecast plot to plot the historical data with forecasts and prediction of intervals. Drawing a Forecast plot on neural network model fit3 using plot() function. > plot(forecast(fit2)) Fig. 4. Forecast Plot on Neural Network 524 Copyright 2017 SERSC

6 To display the summary of the model 2 (Neural network model) > fit2 Series: diff(foodgraints) Model: NNAR(4,2) Call: nnetar(y = diff(foodgraints), maxit = 1000) Average of 20 networks, each of which is a network with 13 weights options were - linear output units. sigma^2 estimated as Conclusion In the study of Time series analysis on agriculture food production data, ARIMA model is used and predicted the values for the next four years. The validity of the predicted values can be checked when the data for the lead periods become available. The model can be used by researchers for forecasting food production in India. However the data need to be updated from time to time with incorporation of current values. References 1. Box, George EP, Gwilym M. Jenkins, Gregory C. Reinsel, and Greta M. Ljung.: Time series analysis: forecasting and control. John Wiley & Sons, Makridakis, Spyros, Michele Hibon, and Claus Moser.: Accuracy of forecasting: An empirical investigation. Journal of the Royal Statistical Society. Series A (General) (1979) 3. Prabakaran, K. and Sivapragasam, C.: Forecasting areas and production of rice in India using ARIMA model. International Journal of Farm Sciences. 4(1), pp (2014) 4. Box, George EP, Gwilym M. Jenkins, Gregory C. Reinsel, and Greta M. Ljung.: Time series analysis: forecasting and control. John Wiley & Sons, Pankratz, Alan.: Forecasting with univariate Box-Jenkins models. Concepts and cases. Vol John Wiley & Sons, 2009 Copyright 2017 SERSC 525