Energy consumption in buildings constitutes a significant portion of global energy use, particularly in hot climates where cooling demands are intensified by climate change and urbanization. Traditional estimation methods often face challenges in capturing dynamic factors such as weather variability, building geometry, and occupant behaviour.

With increasing urbanization and climate change, there is a pressing need for accurate and efficient cooling load predictions in residential settings. Prior research demonstrates that machine learning can significantly improve energy efficiency and operational cost savings, especially in commercial sectors. This study aims to extend those benefits to residential applications by leveraging data-driven approaches.

Despite advances in predictive modeling, current cooling load estimation techniques often lack transparency and adaptability. Furthermore, explainable artificial intelligence (XAI) remains underutilized in this domain, making it difficult for stakeholders to trust and implement AI-based solutions.

How accurately can decision tree algorithms (C4.5, CART, and Decision Stump) predict cooling loads in residential buildings?
Can explainable models improve stakeholder trust and usability in energy forecasting?

To develop a predictive, explainable decision tree–based model for cooling load estimation using thermal and environmental data.
To evaluate and compare the performance of C4.5, CART, and Decision Stump algorithms.
To provide interpretable results that support both residential users and policy decision-makers.

This research contributes an interpretable and efficient modeling approach that enables accurate cooling load forecasting while enhancing transparency through XAI techniques. It supports energy efficiency strategies, reduces operational costs, and aligns with sustainability goals.

The study focuses solely on decision tree (DT) algorithms and their ability to forecast cooling loads in residential buildings. Other machine learning models and broader energy systems are outside the scope. The proposed framework provides actionable insights for homeowners, energy providers, and policymakers to establish more innovative, AI-integrated energy strategies. Future research will address the implementation and evaluation of the developed system.

This study compares three decision tree (DT) algorithms—C4.5, CART, and Decision Stump—to predict cooling loads in residential buildings. While existing literature highlights the effectiveness of decision tree models in energy forecasting, most studies refer to general DT approaches or ensemble methods (such as Random Forest) without directly comparing these specific standalone classifiers. By focusing on interpretable DT algorithms, this study evaluates which model offers the best trade-off between prediction accuracy, computational efficiency, and explainability.

• Data Source: Energy efficiency dataset (768 instances; eight building-related input features such as wall area, roof area, glazing area, and orientation).
• Tools: Standard machine-learning environments supporting classification, feature selection, and evaluation.
• Preprocessing: Data cleaning, normalization, and feature selection to improve model robustness.
• Modeling: C4.5, CART, and Decision Stump trained/tested using a 70/30 split; performance assessed via accuracy, precision, recall, and F1-score.

TABLE I. Key Aspects of Literature Review Summary
This short paper proposes a machine learning–based framework for predicting cooling load in residential buildings using three interpretable decision tree algorithms: C4.5, CART, and Decision Stump. The approach is grounded in the CRISP-DM methodology and leverages publicly available thermal and environmental building data. Although prior literature supports the utility of DT-based forecasting in energy modeling, this study contributes by comparatively analyzing specific DT variants that have not been widely benchmarked together. The proposed methodology supports energy efficiency efforts by enabling transparent and replicable decision-making that can inform residential energy planning and regulatory strategies.

To extend this research, future studies may consider:
• Validation: Apply and test the trained models on real-world datasets across various climate zones and building types.
• Model Enhancement: Integrate DT algorithms with ensemble learning or neural networks to improve performance and scalability.
• Broader Applications: Extend the methodology beyond residential buildings to other sectors—such as commercial real estate and infrastructure—where accurate energy forecasting supports sustainability, cost reduction, and operational planning.

This work serves as a foundation for developing practical tools to support long-term energy sustainability strategies.