Implementation of intelligent learning for energy modelling based on HL for sustainable building

The exchange of indoor and outdoor air through ventilation systems can affect the heating load (HL). Proper ventilation strategies are essential for maintaining indoor air quality. Therefore, this study utilizes the implementation of feature selection approach and Machine learning (ML) techniques for understanding and simulating the behaviour of HL towards various architectural and building behaviours. Five different ML techniques including the Gaussain process regression (GPR), Random Forest (RF), Booted Tree Algorithm (BTA), Support vector machine (SVM), and stepwise regression (SWR) were used in modelling the HL behaviour. Prior to the modelling stage, different pre-analysis techniques were conducted such as dimensionality reduction, correlation analysis and sensitivity analysis. Three different classes were generated based on the input combinations derived from the feature selection method. GPR-C1 demonstrated superior predictive accuracy, achieving a Pearson correlation coefficient (PCC) between 0.995 and 1.000 during training and between 0.994 and 1.000 during validation, thereby outperforming all other models. Calibration models designated C1 produced the best results among the performance measures, especially GPR-C1, BT-C1, and SWR-C1, which showed almost perfect R 2 and few mistakes. Variant C2 produced moderate results across models, preserving respectable error measurements and high R 2 scores. In contrast, variation C3—specifically SVM-C3—produced the lowest results, with low R 2 values, high MSE, and raised RMSE, indicating weaker prediction ability that necessitates more optimization and refining. Furthermore. GPR-C1 display the highest performance skills than all the model combinations used in the current work.