What question did this study set out to answer?

This work aims to develop a drift-aware MLOps pipeline for effective household electricity load forecasting.

April 25, 2026Open Access

Drift-Aware MLOps Pipeline for Household Electricity Load Forecasting

Key Points

This work aims to develop a drift-aware MLOps pipeline for effective household electricity load forecasting.
Integrated XGBoost-based prediction model with drift detection techniques.
Evaluated four retraining strategies across five simulated drift scenarios with 200 total runs.
Utilized Population Stability Index and Kolmogorov-Smirnov test for performance monitoring.
Frequent retraining can degrade model performance, with the no-retraining policy achieving the lowest average MAPE at 42.11%.
Robust initial training is crucial for effective forecasting.
Cautions against unnecessary retraining in low-drift environments.

Abstract

This work presents a drift-aware MLOps pipeline for household electricity load forecasting, integrating an XGBoost-based prediction model with multiple drift detection techniques including Population Stability Index (PSI), Kolmogorov-Smirnov (KS) test, and performance monitoring. We evaluate four retraining strategies across five simulated drift scenarios using rigorous statistical experimentation (200 total runs). Contrary to common assumptions, results show that frequent retraining can degrade model performance, with the no-retraining policy achieving the lowest average MAPE (42.11%). The study highlights the importance of robust initial training and cautions against unnecessary retraining in low-drift environments. This work contributes practical insights for deploying reliable time-series forecasting systems in production MLOps settings.

Drift-Aware MLOps Pipeline for Household Electricity Load Forecasting

Key Points

Abstract

Cite This Study