Rainfall Prediction in Medan City Using Support Vector Machine–Based Machine Learning Models

Abstract

Accurate rainfall classification remains a persistent challenge because precipitation is governed by highly nonlinear atmospheric processes, while routine meteorological observations are often limited, particularly in humid tropical environments. Although recent advances in machine learning have substantially improved rainfall prediction, relatively few studies have examined the capability of interpretable models using compact monthly meteorological datasets. This study developed and evaluated a linear Support Vector Machine (SVM) for classifying monthly rainfall in Medan City using air temperature, relative humidity, sunshine duration, and wind speed as predictor variables. The analysis was conducted using 23 monthly observations collected by BBMKG Region I Medan from July 2024 to May 2026. Rainfall was categorized using the sample mean threshold of 274.7 mm, producing seven higher-rainfall and sixteen lower-rainfall observations. Predictor variables were normalized using the Min–Max method, and the linear SVM parameters were estimated through Sequential Minimal Optimization. The resulting model produced an accuracy of 70%, precision of 50%, recall of 43%, and an F1-score of 46%, with optimization identifying two support vectors and a decision boundary characterized by positive coefficients for relative humidity and wind speed and negative coefficients for air temperature and sunshine duration. These findings indicate that the proposed model successfully identified meaningful rainfall-classification patterns while exhibiting greater capability in recognizing lower-rainfall than higher-rainfall conditions. The study contributes an interpretable machine-learning framework for rainfall classification under limited observational conditions and demonstrates the importance of balancing predictive performance with model transparency in tropical hydrometeorological applications.