Predicting Energy Demand with Interpretable Machine Learning Models

Accurate energy demand forecasting is crucial for planning for a sustainable future. Various atmospheric conditions like temperature, humidity, wind speed, visibility and many other meteorological factors can significantly influence the demand for energy. Machine learning (ML) models utilize sophisticated algorithms to identify and interpret complicated patterns. However, the opaque nature of these models makes them difficult to understand. In this paper, we primarily present the use of QLattice, an Explainable Artificial Intelligence (XAI) model, for energy demand prediction. We used temperature and additional meteorological data for the predictions. The performance of QLattice is compared with other traditional models such as Random Forest (RF) and XGBoost. We also analyze how these models (RF and XGBoost) derive outputs by implementing an interpretability tool like Shapely Additive exPlanations (SHAP). This is compared against the inbuilt interpretability feature of QLattice. Results indicate that the QLattice models is able to achieve the highest $R^{2}$ value, MAE, MSE value of $0.993,111$ and 21316 respectively with proper feature engineering and analysis.