Enhanced accuracy for classification modeling with a Decision Tree on rooftop solar PV

Abstract

Rooftop photovoltaic (PV) systems play a crucial role in Indonesia’s decarbonization agenda; however, research on interpretable classification frameworks that integrate roof geometry and meteorological heterogeneity remains limited. Most previous studies have focused on module-level fault detection rather than assessing installation feasibility at the urban scale. Addressing this research gap, the present study proposes a transparent and data-driven approach using a Decision Tree (DT) model enhanced with Grid Search Cross-Validation (GSCV) to classify the feasibility of rooftop PV installations across 20 Indonesian capital cities. Simulation data generated by PVsyst incorporates multiple tilt and azimuth configurations, as well as local weather variables, representing one-, two-, and three-directional roof geometries. The proposed DT–GSCV model is benchmarked against k-NN, Gaussian Naïve Bayes, Logistic Regression, Random Forest, XGBoost, and CatBoost, demonstrating superior generalization and interpretability. Cross-location validation across five rotational subsets confirms stable performance, with an average accuracy of 91.0% ± 0.3 and an F1-score of 0.90, highlighting the model's robustness across diverse climatic zones. Feature importance and SHAP analyses reveal that irradiation and tilt angle are the most influential factors, while temperature and humidity negatively affect feasibility. The novelty of this work lies in developing a reproducible, interpretable machine learning framework that bridges physical PV modeling and data analytics for rooftop system design. This methodology enables rapid, transparent decision support for optimizing rooftop PV deployment across Indonesia's diverse urban and climatic settings.