Predicting Annual Rainfall for the Indian State of Punjab Using Machine Learning Techniques

Abstract

Excess and deficit of rain has always been a concern for an agricultural state like Punjab. Measure to save crop can be taken if the rainfall values are predicted in advance. The current study attempts to solve rainfall prediction problem using machine learning techniques. Current study evaluates 3 machine learning algorithms - KNN, ELM and SVM applied on the rainfall data and other parameters - humidity, wind speed, max temperature and min temperature for the data from 1973 to 2019. These algorithms were compared on the universal performance parameters - MAE, RMSE, SD, PP and time to predict. We found out that SVM predicts the values nearest to the observed values in the trainer data set and test dataset. The SVM predicted values were not only close to predicted values but also had the least RMSE, MAE and ET. SVM predicted the results with 95% and 92% accuracy for trainer data set and test data set. SVM could have shown better results if number of data points had been more.