Evaluating the impact of data preprocessing to develop a robust MEP-based forecasting model for building integrated with PCM

Data quality is a crucial aspect to accurately predict the energy use of buildings utilizing machine learning methods. Data preprocessing can ensure data quality when a database does not match the criteria for evolving a robust prediction model. Regarding phase change material (PCM)-incorporated buildings, there was no study before this research evaluating the impact of data preprocessing for establishing a robust machine learning based model to forecast their energy consumption (EC). Therefore, for the first time, this research presents an application of the data preprocessing process to compare the results of the formulated multi-expression programming (MEP)-based prediction model's accuracy for predicting the EC of PCM-integrated buildings using processed with actual databases. Data cleaning, outlier detection and removal, and data smoothing were performed on the actual EC database during the data preprocessing process. Results of model evaluation and validation processes for the articulated prediction models showed that the data preprocessing improved the MEP-based prediction model by 33 % to predict the EC precisely. Conclusively, model interpretability (sensitivity, parametric, and energy saving analysis) demonstrated that the developed more reliable prediction model provides energy savings of approximately 20 % by integrating optimum PCM.