Novel Approach Using Random Forest Algorithm for Rainfall Prediction on Imbalanced Dataset

Abstract

Rainfall prediction play a vital role   in the field of agriculture and horticulture and more important to predict rain in desert areas where rainfall very less in rainy season. Many data mining techniques like Ada boosting, Random forest, decision tree, Support Vector Machine (SVM), Neural Network (NN), C5.0, SVR, ANN and K-Nearest Neighbour (KNN) were used to predict various parameters of weather. In this paper we proposed a novel approach based on random forest algorithms, clustering and cross fold validation techniques. The data sets have daily surface data of 3 (three) stations Churu, Jaipur and Jodhpur districts of Rajasthan state, India, form the period 2005 to 2019 (Till September).  The proposed random forest algorithm give highest precision and AUC values for all three locations, precision of Churu is 0.96, Jaipur is 0.89 and Jodhpur is 0.98 for year 2018 rainfall prediction. Ares under curve (AUC) is also higher of proposed random algorithm with compare to other algorithms. The AUC of Churu is 0.85, Jaipur is 0.91and Jodhpur is 0.86 for rainfall in the year 2019.