Beyond explaining: XAI-based Adaptive Learning with SHAP Clustering for Energy Consumption Prediction

Beyond explaining: XAI-based Adaptive Learning with SHAP Clustering for Energy Consumption Prediction


In today’s sustainability-conscious world, accurate prediction of energy consumption in buildings is a matter of crucial importance. This task is made even more challenging by unpredictable events like the global COVID-19 pandemic, which can significantly alter usage patterns. Recognizing these complexities, we have embarked on an exciting project aimed at creating a more adaptable and efficient prediction model— the SHAP Clustering-based Adaptive Learning (SCAL) framework.

Our SCAL framework focuses on forecasting energy consumption in buildings, a topic with far-reaching implications for environmental sustainability and cost efficiency. The uniqueness of the SCAL framework lies in its adaptability—it can dynamically adjust to changing data patterns, a feature that’s particularly useful when dealing with real-world data that’s subject to sudden shifts, like those experienced during the COVID-19 pandemic.

Central to our framework is the use of SHapley Additive exPlanations (SHAP), a method under the umbrella of Explainable Artificial Intelligence (XAI). SHAP has distinctive properties that allow for an in-depth understanding of how each feature contributes to individual predictions. Particularly, it can handle missing data—a common hurdle in real-world datasets—ensuring that our analysis remains robust even in the face of incomplete information.

As part of our project, we are applying the SCAL framework to energy consumption data from 36 buildings, collected from January 2019 to June 2022. This period, marked by the advent of the COVID-19 pandemic, offers a unique testing ground for the SCAL framework’s adaptability to dramatic and sudden changes in energy consumption patterns.

Our work represents a significant step in the field of machine learning, particularly in the context of energy consumption forecasting. By incorporating the SCAL framework, we aim to enhance the adaptability and interpretability of predictive models—both critical factors when deploying machine learning models in real-world scenarios.

We are currently in the midst of this exciting project, in partnership with Software AG and Granlund. As we delve deeper into this research, we warmly invite students, scholars, and fellow researchers to join us on this journey. We welcome ideas, critiques, and contributions to help shape this project and further advance the dynamic and crucial field of adaptive and explainable machine learning.

Partner: Software AG, Granlund

Project Status: Ongoing