Analysis of Factors Affecting the CSI300 Index Based on KPCA and Various Machine Learning Algorithms

Abstract

This paper utilises Kernel Principal Component Analysis (KPCA) and various machine learning algorithms to analyze the importance of factors affecting the Chinese Securities Index 300 (CSI300). Based on previous research, this paper constructs an indicator system consisting of 4 secondary and 21 tertiary indicators affecting the CSI300. The data is then reduced through KPCA and processed by various machine learning algorithms, including LightGBM, XGBoost, SVM, and Random Forest, to compare their predictive ability and feature importance. The results indicate that: (1) Under appropriate model parameter settings, the LightGBM model performs the best, while the other algorithms, such as the XGBoost, SVM, and Random Forest models, perform worse and with greater variability than the former. (2) This paper identifies the most significant indicator factors that affect the CSI300 index, such as closing price, price-to-book ratio, and turnover rate. Conversely, some factors, such as the buy-to-sell ratio, exhibit lower importance. These research findings have certain reference and guiding significance for improving the accuracy and reliability of stock market forecasting and practical and theoretical research in financial markets.