Rainfall-Based Crop Selection Model Using MapReduce-Based Hybrid Holt Winters Algorithm

Abstract

Water is essential for all life processes, cannot be an alternative, and is used for transportation; it is a source of energy and serves many useful purposes like agriculture and industry-related activities. Rain is the major source of irrigation in the India, and it is directly connected with the crop production. Yields of crops, especially in rain-fed region, depend on rainfall patterns. Therefore, it is very important to predict rainfall quantity from historical rainfall data using statistical methods [1]. Another important factor is soil moisture that purely depends on the rainfall. It also play vital role in the crop production. Therefore, rainfall prediction is the most important area and needy tool. It has very good impact in determining the food production. The most popular and frequently used data mining algorithms are K-means, C4.5, SVM, EM, Apriori, AdaBoost, PageRank, Naive Bayes, KNN, and CART. The authors of [2] provided detailed explanations of each of these methods and discussed their impact on current and future research in these methods. The rainfall forecasting for boyolai, central java, Indonesia is made by using the Z-Score model and Winters triple exponential smoothing. The authors predicted the weather based on climate classification using the Oldman method [3]. The weather-based crop yield prediction model was proposed in [4] to forecast the yield of 12 major California crops. But very few crop yields were predicted with high accuracy, suggesting that crop yield variability should be captured for more accurate forecasting. The time series analysis using HW algorithm is used to identify level, trend, and seasonality in the rainfall dataset and forecast the districtwise rainfall in Tamilnadu, India, based on these information [5]. In this paper, forecasting the districtwise rainfall using hybridization of KNN with MapReduce-based HW. MapReduce framework is used here for the effective initialization process of HW algorithm iteratively. The experimental study is designed with different initializa-