Enhancing forecasting accuracy in dynamic environments via PELT-driven drift detection and model adaptation
Nikhil Pawar, Guilherme Vieira Hollweg, Akhtar Hussain, Wencong Su, Van‐Hai Bui · Next Energy · 2025
Time series forecasting models often experience a decline in prediction accuracy due to data drift, which occurs when the underlying data distribution changes over time. To address this challenge, this study proposes an adaptive forecasting framework that integrates drift detection with targeted model retraining to compensate for drift effects. The framework utilizes the Pruned Exact Linear Time (PELT) algorithm to identify drift points within the feature space of time series data. Once drift intervals are detected, selective retraining is applied to prediction models using Multilayer Perceptron and Lasso Regressor architectures, allowing the models to adjust to changing data patterns. To assess effectiveness, the method is applied to a synthetic dataset for ideal conditions and a real-world heating, ventilation, and air-conditioning dataset that reflects practical challenges and complex dependencies. Initial baseline models were developed without drift detection using extensive feature engineering. After integrating drift-aware retraining, the multilayer perceptron (MLP) model achieved a 27% reduction in Mean Absolute Error and a 4–5% increase in R ² on the real-world dataset, while even greater improvements were observed on the synthetic dataset. Similar enhancements were achieved with the Lasso Regressor. These results highlight the robustness and generalizability of incorporating drift detection and adaptive retraining to sustain forecasting accuracy across diverse domains.