A Comparative Machine Learning Framework for Stock Market Forecasting and Dynamic Portfolio Optimization

Predicting stock market trends has always been an important area of research, especially in the application of machine learning algorithms. This study aims to predict stock returns in the steel, electronics, food and beverage, and real estate industries using three different machine learning models: linear regression, random forest, and gradient boosting. The research utilizes historical data from 2005 to 2024, including key indicators such as MACD, RSI, and price-to-earnings ratio (PE/TTM), to forecast stock returns. Additionally, the study implements a portfolio optimization model based on the Markowitz model, incorporating different turnover limit to simulate a real trading environment and optimize asset allocation. The results indicate that as turnover limit increase, portfolios with dynamic rebalancing perform overall better across industries compared to others, reflecting the critical role of total fee rate management and turnover limits in maximizing portfolio returns. This study emphasizes the value of combining machine learning models with portfolio optimization strategies to enhance prediction accuracy and asset allocation efficiency.