SCD-CA: Stable Concept Drift Learning Based on Correlation Alignment for Streaming Data Forecasting
Ge Yu, Zixiao Zhang, Xi Zhang · 2022 13th Asian Control Conference (ASCC) · 2022
In various real-world cases, the streaming data is sequentially collected over time. It means the distributions could change due to the non-stationary and uncertain environments over time, which is, the concept drift. To handle concept drift, in this paper, we propose a novel concept drift adaptation method SCD-CA, which conducts the adaption models from historical data to forecast the future unseen data. Our method can take a correlation alignment to alleviate the bias caused by fake correlations among variables. In particular, we design an analyzable optimization process to find optimal parameters for unseen streaming data forecasting. Experimental results validate the superiority of our method in wind field dataset, showing it is feasible and promising.