Optimizing Climate Forecasts Across 16 Zones Using Regression-Based Machine Learning Models

Abstract

The XYZ Climatology Station faces challenges in improving the accuracy of decadal rainfall forecasts, with an average achievement of 57.4% in 2022 and 58.8% in 2023, below the organizational performance target of 70% accuracy as set in its strategic objectives. This study aims to develop machine learning-based predictive models for 16 climate zones to enhance forecast accuracy. Five regression algorithms—Multiple Linear Regression, Support Vector Regression, Extra Trees Regression, Random Forest Regression, and Decision Tree Regression—were tested under two scenarios: input variable variations (VR) and time series data length (TS). Results showed that the VR scenario increased average accuracy to 71.7% (2022) and 69.4% (2023), while the TS scenario achieved 73.1% (2022) and 72.6% (2023). Support Vector Regression and Extra Trees Regression demonstrated the best performance in most zones. These models are expected to be operationalized to improve climatological information services and better meet public and stakeholder needs.