Machine Learning-based Financial Market Volatility Prediction Models

Aiming at the problem that traditional technical analysis is difficult to accurately capture the non-linear fluctuation of stock prices, this paper takes stock 300059 as the research object, and based on the historical trading data from 1st April to 30th August, 2025, constructs the Random Forest Dual Model (Classification model predicts the direction of upward and downward movement, and Regression model predicts the closing price) and the multi-dimensional technical indicators such as moving average line, price fluctuation and RSI are extracted as input. The feature correlation is verified through visual analysis, and the model is compared with linear regression and support vector regression. The results show that the random forest regression model has MAE of 0.194, MSE of 0.088, an R2 of 0.981, and a classification model accuracy of 91.8%, which is significantly better than the comparative model; the feature importance analysis shows that the closing price and the intraday high and low prices contribute the most.