A Random Forest Stock Prediction Model Based on Bayesian Optimization

Random forest models often rely on human-determined parameters, which may introduce subjectivity and affect the degree of optimization of the model. To address this issue, we use a random forest approach based on Bayesian optimization to reduce the influence of human factors and improve the fitting. Bayesian optimization prevents the model from falling into the dilemma of local optimum. In order to show the effect of the optimization, we compare the support vector machine, the original random forest and the LGBM model, and perform Bayesian optimization on the latter two to achieve better results. Experiments show that the optimized model performs well in stock return prediction with a reduction of 0.01 in RMSE and MAE, a reduction of 0.4 in MAPE, and an improvement of 0.02 in R-2 value compared to the other models. The Bayesian optimization based Random Forest model achieves the lowest prediction error on different datasets and has wide applicability.