Index Prediction Model Based on LASSO-PCA and Deep Learning

Predicting financial market trends poses significant challenges due to the complex, dynamic, and often chaotic nature of the market, especially when dealing with data featuring a multitude of characteristics. In this research, we propose an effective data mining approach that combines Least Absolute Shrinkage and Selection Operator (LASSO) and Principal Component Analysis (PCA) for two-stage feature dimensionality reduction, resulting in a refined dataset. To enhance the model's capacity to capture medium- and long-term index trends, we implement a sliding time window approach, utilizing data from the preceding 60 trading days. Long-Short Term Memory (LSTM) and Gated Recurrent Unit (GRU) models are employed to predict the return rate of Securities Bank of China (399986) over a 30-day trading period. We conducted a comprehensive comparative analysis involving our proposed model and established methods, namely, LASSO, PCA, and Hybrid Recurrent Neural Networks (RNN). Our empirical findings unequivocally demonstrate the superior performance of our model in terms of both prediction accuracy and stability. Specifically, our model exhibits significantly higher predictive accuracy when forecasting the return rate of Securities Bank of China (399986) over a 30-day trading period, all while maintaining enhanced stability. These results underscore the exceptional efficacy of our approach within the realm of financial market time series forecasting, thus providing robust support for further research and practical applications within this domain.