Predicting Next-Day Rainfall Using Machine Learning Techniques

Predicting rainfall using physical models is inherently complex due to involvement of a large number of variables. Accurate and timely predictions of rainfall have the potential to mitigate human and financial losses. Machine learning techniques capture non-linear relationships between variables more effectively than traditional statistical methods. The major objective of the present research is to develop machine learning (ML) models based on logistic regression, decision trees, random forests, and artificial neural networks (ANNs) for predicting next-day rainfall using a comprehensive set of environmental variables. The effectiveness of these ML models was tested using an extensive dataset that comprises of 145460 instances containing 21 quantitative variables from 49 weather stations in Australia. Evaluation of model efficacy was conducted employing an array of performance metrics based on accuracy, precision, recall, and F1 score. The findings indicate that the random forest model demonstrated superior performance compared to other techniques, with the ANN model exhibiting performance that was closely comparable. Random forest achieved the highest accuracy of 85.55% among the models considered and demonstrated strong precision (75.28%) along with a well-balanced F1-score of 56.11%. It can be concluded that the random forest model is an excellent choice for applications where both accuracy and robustness are essential.