کنفرانس بین المللی فناوری اطلاعات و دانش

صفحه اصلی / شانزدهمین کنفرانس بین المللی فناوری اطلاعات و دانش

Beyond One-Hot: CatBoost for Heating and Cooling Load Prediction

نویسندگان :

Shayan Naghizadeh¹ Mohammad Saeed Rajabi² Ehsan Nazerfard³

1- Amirkabir University of Technology 2- Amirkabir University of Technology 3- Amirkabir University of Technology

کلمات کلیدی :

Building energy efficiency،load prediction،CatBoost،boosting algorithms،machine learning

چکیده :

This Accurate prediction of heating and cooling loads in buildings is essential for energy-efficient design and operation. This study benchmarks multiple machine learning regression models on the ENB2012 dataset and evaluates the impact of encoding strategies for categorical-like features. Traditional regression and ensemble models were compared against CatBoost, a gradient boosting algorithm with native support for categorical variables through ordered target statistics and ordered boosting. CatBoost achieved the lowest prediction errors, with reductions of up to 30% in MAE and RMSE compared to XGBoost and the Random Forest Regression model, while maintaining R2 above 0.99 for both heating and cooling loads. Feature-importance analysis highlighted compactness, overall height, and surface area as dominant predictors, aligning with physical principles of heat transfer. These findings demonstrate that CatBoost provides a robust and interpretable framework for accurate building-load forecasting beyond one-hot encoding.