Machine Learning Application in Stock Portfolio Optimization and Price Prediction

This article selects the top ten stocks with the highest returns in the CSI A50 index, and first uses the wavelet denoising method to smooth the original stock price data. Compared with VaR, this article uses CVaR to reduce the potential impact of extreme events, which can better reflect tail risk. Based on this, a high return, low-risk investment portfolio was constructed, and three machine learning models LSTM, SVM, and XGBoost were used to predict future stock prices. After comparative analysis, the XGBoost model with the highest prediction accuracy was selected as the final model. The research findings can provide investors with a new investment strategy by combining wavelet analysis, risk management, and machine learning prediction to optimize portfolio construction and predict future returns.