Weighted Hybrid Machine Learning Algorithm to predict Rainfall for Hyderabad Region

A Weighted Hybrid Machine Learning algorithm using Classification cum regression for the predicting the amount of rainfall in Hyderabad region is proposed. The classification models are used for predicting the occurrence of rainfall, and regression models are used for predicting the value of Rainfall in mm. The methodology involved the weighted average of two hybrid algorithm that estimates the occurrence of rain and the amount of rainfall over Hyderabad Region. Five-year Data was collected from Hyderabad Metrological station along with few related features responsible for rainfall over particular region. Experimental results reveal that the predictions using the proposed approach are consistently better than those obtained using the other methods. The Predicted data for rainfall is later compared with Real time actual data of rainfall for April and May 2020 and found to yield 75% accuracy in the above said period.

I. Introduction

The Primary occupation of India is agriculture. The Crop Seasons of India are referred to Rabi, Kharif and Zaid. Kharif season is the rainy season which spans from June to September. Hence Rainfall forecasting is an essential part of farmer's routine. Over the years many algorithms had been formulated for rainfall prediction few of them are listed below. Tarun et al found that Artificial Neural Networks handled a large data with inconsistency with an accuracy of 87% [1]. Shah et al proposed a model that identifies minimum and maximum temperature, Wind Speed and humidity using Auto Regressive Integrated Moving Average (ARIMA) and Neural Network methods [2]. Singh and Kumar identified a combined algorithm using Random Forest and Gradient boosting that had a better F1 score to predict rainfall [3]. Rodrigues and Deshpande introduced Multiple Linear Regression for the Rainfall statewide in a monthly basis. The authors used rainfall values of 2014 to predict the rainfall [4]. Manek and Singh used data from India Water portal which had 1901–2002 that comprises of Precipitation, Cloud Cover, Vapour pressure and Average temperature. The Authors used MATLAB neural Network for simulation of different models. The Radial basis function neural Network performed better on the above data than others [5]. Geetha and Nasira developed a Rapid miner that involves 12 parameters that brings out weather warnings which acted as disaster prevention tool for farmers. It achieved an accuracy of 80.67% in the year 2014 [6]. Tirumalai et al used yester year data on linear regression method using dependent and independent variable for the prediction of crop yield [7]. Kala and Vaidyanathan proposed a model that learned functional relationship of original and predicted data. The Authors used feed forward neural network for the prediction of rainfall which had an output of 93.55% [8]. Naveen and Mohan defined a prediction model which utilizes information including different fields in agriculture, water system used, solar power based power back up etc. The model used Wavelet-based SVM and Radial Basis Function Neural Network (RBFNN) utilizing Hybrid Particle Swarm Optimization using Genetic Algorithm (HPSOGA) which yielded wonderful prediction results [9].