Segment-based ensemble learning model for stock price forecasting

Stock price forecasting is sometimes affected by factors such as timeliness and nonlinearity. Therefore, using traditional forecasting methods to model the entire dataset often results in problems such as low forecasting accuracy. These problems are collectively referred to as conceptual drift problems when the stock data does not match the degree of change in the previous period due to intra-flrm changes. In this paper, we propose a segmentation modeling and integration learning approach to segment the data set in temporal order and generate subsets of data with the same distribution characteristics, and then integrate them using an integration learning algorithm such as Adaboost to generate several models. The results show that such concept drift problems can be effectively solved by using segmentation modeling and integration learning methods.