RAINFALL PREDICTION FOR MINNA METROPOLIS USING ARTIFICIAL NEURAL NETWORK

Abstract

The effect of rainfall in our society today is stupendous. Rainfall is seen as a benefit to crops and lives. Accurate and timely rainfall prediction can be very helpful for effective security measures for planning water resources management, transportation activities, agricultural tasks, managing flights operations, issuance of flood warning and flood situation. This study aims to predict the rainfall of Minna metropolis. Atmospheric data comprising those of maximum temperature, minimum temperature, relative humidity and rainfall for four consecutive years spanning from January 2015 - December 2018 were acquired from the Geography Department of Federal University of Technology, Minna. The datasets were preprocessed and normalised, and further partition into three parts: 70% for training set, 15% for testing set and 15% for validating set. Feed forward neural network and binary classification was used for the prediction. The target data (rainfall) was labelled as positive or negative (rainfall or no rainfall), that is, (1 or 0) with threshold of 0.5 for classifying the rainfalls. The outcomes of prediction were evaluated using confusion matrix. The best test result indicates that 66 days were predicted to have rainfall and 120 days predicted for no rainfall with 69% accuracy, 1.3% error, 63% sensitivity and 84% specificity. The best validated results also indicate that 77 days were predicted to have rainfall and 109 days predicted for no rainfall with 59% accuracy, 1.4% error, 52% sensitivity and 78% specificity. The performance of the classifier is 0.568 (AUC = 57%).