Disaggregating Household Water End-Uses: A Comparative Study Between XGBoost and TabNet

In a world where water is consumed at an unsustainable level, smart water-saving approaches could play a fundamental role in managing and preserving this resource. By performing water disaggregation, households, corporations and policymakers can identify where water is being used and accordingly come up with water-saving strategies. This paper explores the performance of two classification models for water disaggregation based on XGBoost and TabNet architectures where their explainability is also investigated. A UK case study is then used to validate the proposed approach with water consumption data collected from real homes. The dataset is also augmented with Gaussian noise to test the reliability of the model. The findings suggest that both XGBoost and TabNet water disaggregation models perform similarly well across all water activities while XGBoost maintains the best balance between performance and computational requirements. With the addition of Gaussian noise in the dataset, the XGBoost model is found to be the most reliable model.