Region-Wise Rainfall Prediction Using MapReduce-Based Exponential Smoothing Techniques

Abstract

Weather acts an important role in agriculture. Rainfall is the primary source of water that agriculturist depends on to cultivate their crops. Analyzing the historical data and predicting the future. As the size of the dataset becomes tremendous, the process of extracting useful information by analyzing these data has also become repetitive. To defeat this trouble of extracting information, parallel programming models can be used. Parallel Programming model achieves this by partitioning these large data. MapReduce is one of the parallel programming models. In general, Exponential Smoothing is one of the methods used for forecasting a time series data. Here, data is the sum of truth and error where truth can be “approximated” by averaging out previous data. It is used to forecast time series data when there is Level, Trend, Season, and Irregularity (error). In this paper, Simple Exponential Smoothing, Holt’s Linear, and Holt-Winter’s Exponential Smoothing methods are proposed with MapReduce computing model to predict region-wise rainfall. The experimental study is conducted on two different datasets. The first one is Indian Rainfall dataset which comprises of the year, state, and monthly rainfall in mm. The second is Tamil Nadu state rainfall dataset which consists of the year, districts, and monthly rainfall in mm. To validate these methods, MSE accuracy measure is calculated. From the results, Holt-Winter’s Exponential Smoothing shows the better accuracy for rainfall prediction.