Stock Decision-Making Based on Machine Learning Models and Z-Scores

This study examines the performance of various machine learning models in stock decision-making based on the stock Z-scores. We employed four models—OLS linear regression, decision tree regression, support vector machine regression (SVM), and multilayer perceptron neural network (MLP). By leveraging Z-scores calculated by different models, stock trading strategies were constructed, and the models’ risk-adjusted returns and cumulative returns were assessed. The experimental results indicate that the decision tree model outperformed the others overall, especially in the unweighted case where it showed lower max drawdown and relatively lower risk. After weighting, the performances of the SVM and MLP models improved, with the decision tree regression model achieving the highest cumulative return.