Predicting residential building cooling load with a machine learning random forest approach

Accurately estimating cooling loads in energy-efficient building design is paramount. This abstract delves into a comprehensive study of the prediction of cooling loads in residential buildings by applying a Random Forest model. Additionally, the study introduces a novel dimension by optimizing this model using two cutting-edge metaheuristic algorithms: Tasmanian Devil Optimization and Northern Goshawk Optimization. The initial goal of this research involves developing a robust RF model capable of predicting CL in residential buildings. Leveraging a vast dataset comprising architectural and environmental parameters, this model is trained to make precise predictions, thereby enhancing the overall energy efficiency of buildings. TDO and NGO iteratively search for optimal hyperparameters and model configurations, ultimately improving accuracy and generalization. This study's results demonstrate the RF model's effectiveness in predicting cooling loads with a high degree of accuracy. Furthermore, the incorporation of TDO and NGO optimization techniques significantly enhances the model's performance, reducing energy consumption and costs associated with cooling systems in residential buildings. Specifically, the RFTD model stands out as it consistently delivers the most precise and dependable results compared to other models. This superiority is exemplified by a remarkable R² value of 0.997 and a notably low RMSE value of 0.498. This research provides a pioneering approach to the estimation of CL in residential buildings. Combining machine learning (ML) with the ingenuity of metaheuristic optimization offers a promising avenue for achieving sustainable and energy-efficient building design, aligning with contemporary environmental goals and economic imperatives.