Accurate rainfall prediction is essential for managing water resources, mitigating disasters, and supporting sustainable agriculture in the face of increasing climate variability. However, the high dimensionality and heterogeneity of meteorological data often hinder the efficiency and interpretability of traditional models. This study introduces a feature selection driven machine learning framework that enhances rainfall classification accuracy and computational sustainability using Australia's comprehensive weather dataset. Three classifiers Logistic Regression, Random Forest, and a Multi-Layer Perceptron (MLP) neural network were evaluated before and after applying the feature selection. The proposed framework demonstrates that dimensionality reduction significantly improves efficiency while preserving or enhancing predictive capability. Experimental results show up to 96 % reduction in training time and a perfect ROC-AUC score (1.00) for both Random Forest and Neural Network models, with false negatives reduced by 30–68 %, leading to more reliable rain-event detection. The study's novelty lies in systematically quantifying the effect of feature selection on model robustness and efficiency for rainfall prediction a topic that has received limited attention in prior research. These findings provide a sustainable and generalizable approach for real-time rainfall forecasting and can be extended to other climate prediction tasks requiring efficient model deployment under computational constraints.
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号