Interpretable Rainfall Forecasting Using SHAP-Enhanced Machine Learning: A Case Study on U.S. Urban Climate Data (2024–2025)
Accurate rainfall prediction supports resilient communities, sustainable agriculture, water distribution, and disaster preparedness. Because rainfall depends nonlinearly on interacting meteorological conditions, conventional forecast-ing methods can be inadequate for short-term, localized events. This study develops a machine-learning framework for next-day rainfall classification using daily weather observations from 20 major U.S. cities during 2024–2025. The dataset includes temperature, relative humidity, wind speed, precipitation, cloud cover, and pressure, together with a binary Rain Tomorrow label. Six classifiers are evaluated: Random Forest, Support Vector Machine (RBF kernel), k-Nearest Neighbors, Logistic Regression, Naive Bayes, and Linear SVM. SHapley Additive exPlanations (SHAP) are used to interpret model predictions and feature contributions. Random Forest reports the strongest results, with precision, recall, and F1-score of 1.00; RBF SVM and KNN achieve F1-scores of 0.97 and 0.94, respectively. Logistic Regression, Linear SVM, and Naive Bayes achieve F1-scores between 0.76 and 0.77. SHAP analysis identifies precipitation and humidity as the most influential variables, consistent with meteorological expectations. These results suggest that combining an ensemble classifier with interpretable explanations can support transparent rainfall forecasting for agriculture, urban planning, and emergency response.
Volume & Issue
Vol. Volume 9 / Iss. Issue 2