Introduction: Predicting stock returns, although complex, has always been of interest to investors. Investors and decision makers need information to predict stock returns. Aim: The aim of this study was to predict stock returns in companies listed on the Tehran Stock Exchange by emphasizing the role of financial and regulatory criteria using machine learning methods. Method: This research is applied in terms of purpose and in terms of the type of field-library study. The statistical population consisted of companies listed on the Tehran Stock Exchange during the years 1390 to 1396 (a seven-year period) and the hypotheses were tested using machine learning methods (linear, nonlinear and cart). Results: The results of variable selection test using Relief-F method indicate a significant relationship between return on assets, price-to-earnings-per-share ratio, quota, operating cash ratio, financial leverage and institutional owners' ratio to stock returns. The results also showed that PINSVR machine learning methods in linear and nonlinear mode and CART have a good ability (more than 90%) to predict stock returns, but the nonlinear PINSVR method and CART showed higher predictive power than the linear method.
Conclusion: According to the results, capital owners and company decision makers are advised to use the predictive power of artificial intelligence algorithms, especially nonlinear methods, in their decisions about investing in the Tehran Stock Exchange.