Optimizing Machine Learning for Imbalanced Rainfall Data Prediction

Unpredictable weather changes are closely related to extreme weather and rainfall. Extreme weather that occurs due to drastic atmospheric changes, such as very high rainfall, is the main trigger for hydrometeorological disasters, namely floods, landslides, strong winds, and droughts. This creates obstacles in various fields of human activities, such as transportation, agriculture, and outdoor activities. Therefore, the ability to predict accurate rainfall is necessary to increase awareness of potential hazards. The complexity of rainfall data makes it difficult to forecast when the data has imbalanced characteristics, as they can affect forecasting performance. Machine learning approaches have shown promising results in improving rainfall prediction accuracy by capturing complex patterns between various meteorological factors and rainfall patterns. This research is intended to evaluate the performance of several machine learning algorithms, namely Decision Tree, Support Vector Machine (SVM), Naive Bayes (NB), and Logistic Regression, in rainfall prediction. The test results show that the Naive Bayes model provides outstanding results with an accuracy rate of $93.9 \%$, outperforming the other algorithms in predicting rainfall. Meanwhile, Decision Tree recorded 83.7% accuracy followed by SVM 83.6% and Logistic Regression $89.2 \%$. This research shows that machine learning approaches, particularly Naive Bayes, are effective in predicting rainfall and can contribute to hydrometeorological disaster mitigation efforts.