Interpretable Machine Learning for Time Series Forecasting: The Case of Natural Gas Consumption

The thesis addresses the challenge of predicting monthly Natural Gas consumption to mitigate the issue of irregular measurements faced by energy resellers, particularly for residential users. The goal is to develop accurate predictive models based on time series data. To this end, the study proposes a machine learning-based approach that adapts traditional regression models to time series forecasting using a supervised sliding window technique. This transformation allows high-performance tabular models to be applied together with Explainable AI (XAI) tools to make forecasts not only accurate but also interpretable. By improving transparency for energy suppliers and providing end users with a clearer view of their consumption behavior, the proposed solution contributes to more efficient, transparent and sustainable energy management, especially in the evolving context of smart meters and market liberalization.