Cluster-Based Machine Learning Approaches for Predicting Daily Maximum Temperatures in Indonesia Under Climate Change

Climate change is increasing the frequency of extreme temperatures in Indonesia, creating significant prediction challenges due to its geographical diversity. To address this, the study proposes a spatially adaptive framework using BNU-ESM and ERA5 data (1980–2005). The Indonesian region was classified into four climate clusters via K-Means, where Support Vector Regression (SVR), Random Forest (RF), and XGBoost models were evaluated. Results show SVR consistently outperformed other models across all clusters. In stable regions, SVR achieved the highest accuracy (RMSE 0.10; MAE 0.08) and remained superior even in the most volatile clusters. The study's novelty is the integration of clustering with comparative model evaluation, offering a robust methodology for precise, regionally adaptive climate early warning systems.Practically, this predictive model can support national mitigation strategies by enabling proactive resource allocation and targeted interventions in high-risk climate zones.