Machine Learning-based Rainfall Prediction from Weather Data: A Comparative Analysis

Rainfall forecasting has drawn the greatest interest in recent years due to its complexity and the importance of detecting the possibility of flooding, monitoring pollutant contamination, water supply and agricultural usage. Existing instrument-based and statistic models used in the meteorological department are old-fashioned and expensive. Due to the high volume of weather-related data captured by Rader and billions of sensors, inaccurate statistic models, and the hysteretic and reactive nature of instrument-based models, machine-learning (ML)-based forecasting has attracted huge attention in rainfall prediction. These ML models include different supervised ML models such as Linear Regression, Random Forest and Support Vector Machine (SVM). However, few existing works have examined the advantages and disadvantages of diverse ML models to study their efficiency in the domain of rainfall prediction. Therefore, this work proposes a comprehensive analysis of different ML models for accurate and fast rainfall prediction. In this work, we develop several ML models, namely linear regression, linear regression with parameter penalty, cross-validation with linear regression, linear regression with principal component analysis (PCA), support vector regression (SVR) without PCA, support vector regression (SVR) with PCA, and finally artificial neural networks (ANN). We train those models with climate data from Colorado, a significant American metropolis, for the years 2015 to 2018 and compare their performance (prediction accuracy) in terms of validation loss. The results demonstrate that the proposed artificial neural network achieves the best accuracy.