Benchmarking Six Machine Learning Algorithms for Accurate Precipitation Forecasting

Predicting rainfall is essential for agriculture, managing water resources, and even in disaster planning. Good rainfall prediction mitigates economic impacts and saves lives. Traditional methods for rainfall forecasting using weather data are often complicated. They fail to achieve sufficiently reliable accuracy due to variability and complexity; as a result, more advanced techniques that learns from historical data are needed. In this paper, we assess the performance of rainfall prediction using datasets from two different sources and locations with six selected models: Random Forest (RF), Decision Tree (DT), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Logistic Regression (LR) and Neural Network (NN). On Dataset 1, the RF and DT methods achieved perfect accuracy (100%) but did not perform well on Dataset 2, showing a drop to $78-82 {\%}$ accuracy which demonstrates their dependence on the dataset. Neural Networks (NN) exceeded expectation on Dataset 2 and almost perfectly compared to SVM, who was consistent across both datasets with 84.7-95.8 It was unexpected that LR and KNN would yield such poor performance results when juxtaposed with the ensemble and deep learning models. These results indicate that tree-based models (RF, DT) only achieved 100% accuracy on selected datasets but the “best” model demonstrated a clear over-reliance on neural networks Multi-Layer Perceptrons for the other datasets, suggesting the need for more scrutiny. In addition, the model performance on different datasets is context dependent which reveals the need for tailored algorithm selection in specific contexts. To summarize, the current study highlights the growing impact of artificial intelligence in the field of meteorology where it advocates for the use of neural networks and SVM for dependable rainfall forecasting. Further studies should focus on the use of combined models and increased multi-source datasets to better the generalization and practicality of the models.