Fig 1.
Architecture of explainable residential building electrical energy consumption modeling.
Table 1.
Statistical values and key attributes for residential building energy datasets.
Table 2.
Selected input variables, descriptions, and types.
Table 3.
Advantages and disadvantages of decision tree–based ensemble learning methods.
Table 4.
List of hyperparameters for decision tree-based ensemble learning method.
Fig 2.
Taylor diagram showing the performance of ensemble learning models on the University Residential Complex dataset.
Fig 3.
Taylor diagram showing the performance of ensemble learning models on the Appliances Energy Prediction dataset.
Table 5.
Comparative analysis of decision tree–based ensemble learning models trained with external factors on the University Residential Complex dataset.
All evaluation metrics are presented in percentage (%).
Table 6.
Comparative analysis of decision tree–based ensemble learning models trained with internal factors on the University Residential Complex dataset.
All evaluation metrics are presented in percentage (%).
Table 7.
Comparative analysis of decision tree–based ensemble learning models trained with external and internal factors on the University Residential Complex dataset.
All evaluation metrics are presented in percentage (%).
Table 8.
Comparative analysis of decision tree–based ensemble learning models trained with external factors on the Appliances Energy Prediction dataset.
All evaluation metrics are presented in percentage (%).
Table 9.
Comparative analysis of decision tree–based ensemble learning models trained with internal factors on the Appliances Energy Prediction dataset.
All evaluation metrics are presented in percentage (%).
Table 10.
Comparative analysis of decision tree–based ensemble learning models trained with external and internal factors on the Appliances Energy Prediction dataset.
All evaluation metrics are presented in percentage (%).
Table 11.
Harmonic mean comparison considering various input variable configurations on the University Residential Complex dataset, presented as percentages (%).
Table 12.
Harmonic mean comparison considering various input variable configurations on the Appliances Energy Prediction dataset, presented as percentages (%).
Table 13.
Performance comparison between deep learning models and the GBM model on the University Residential Complex dataset.
All evaluation metrics are presented in percentage (%).
Table 14.
Performance comparison between deep learning models and the CatBoost model on the Appliances Energy Prediction dataset.
All evaluation metrics are presented in percentage (%).
Fig 4.
Influence of input variables in the gradient boosting machine (GBM) model on the University Residential Complex dataset: (a) feature importance; (b) Shapley additive explanations (SHAP) summary plot.
See Table 2 for abbreviation definitions.
Fig 5.
SHAP summary plots of other ensemble learning models on the University Residential Complex dataset: (a) random forest (RF); (b) extreme gradient boosting (XGBoost); (c) light gradient boosting machine (LightGBM); (d) categorical boosting (CatBoost).
See Table 2 for abbreviation definitions.
Fig 6.
Comparison of partial dependence plots (PDPs) from the GBM model on the University Residential Complex dataset: Analyzing the relationship between holiday status (Holi) and average electricity consumption (Cons_avg) (a) using scikit-learn; (b) using SHAP analysis.
See Table 2 for abbreviation definitions.
Fig 7.
Comparison of PDPs from the GBM model on the University Residential Complex dataset: Analyzing the relationship between temperature-humidity index (THI) and Cons_avg (a) using scikit-learn; (b) using SHAP analysis.
See Table 2 for abbreviation definitions.
Fig 8.
Influence of input variables in the CatBoost model on the Appliances Energy Prediction dataset: (a) feature importance; (b) SHAP summary plot.
See Table 2 for abbreviation definitions.
Fig 9.
SHAP summary plots of other ensemble learning models on the Appliances Energy Prediction dataset: (a) RF; (b) GBM; (c) XGBoost; (d) LightGBM.
See Table 2 for abbreviation definitions.
Fig 10.
PDPs of the CatBoost model on the Appliances Energy Prediction dataset: (a) Hour_x and Hour_y; (b) holiday indicator (Holi) and Hour_y.
See Table 2 for abbreviation definitions.
Fig 11.
SHAP decision plot for the test set on the Appliances Energy Prediction dataset.
See Table 2 for abbreviation definitions.
Fig 12.
Scatterplots and R2 comparison for each decision tree–ensemble learning model.
(a) random forest; (b) gradient boosting machine; (c) extreme gradient boosting; (d) light gradient boosting machine; (e) categorical boosting; (f) R2 comparison.
Fig 13.
Summary plot of the RF model on the hourly Appliances Energy Prediction raw dataset.