The stock market is commonly influenced by a variety of factors, and the dependency structure between stock market and economic factors can also change. This paper utilizes Principal Component Analysis for factor analysis on the time series dataset of a single stock, compares the dimension reduction effects across different time windows, and employs a PCA-BP hybrid model to predict the closing price of the stock within a specified time window and assess its performance. The results indicate that for a segment of time series data, the smaller the time window, the relatively better the effect of PCA is. In experiments, using the most recent 6 months of data showed better PCA dimension reduction and model prediction effects compared to 12 and 60 months. Therefore, in the principal component analysis and prediction of single stock, an overly long time span may not necessarily bring better returns. Instead, data and a richer quantity of factors that are closer to the predicted time window should be selected to meet the learning requirements of complex predictive models.