The Comparative Assessment of Data Mining Methods Effectiveness to Forecasting Return and Risk of Stock in Companies Listed in Tehran Stock Exchange

In this study, with using of four algorithms: linear discriminant analysis algorithm, quadratic discriminant analysis algorithm, K-nearest neighbors algorithm and decision tree with the help of 16 independent variables has been addressing to predict stock returns and systematic risk. Four algorithms are running once with using of whole independent variables and one again with using of 4 independent variables that are known with using of filtering approach as the most effectiveness of variables in predicting the return and risk. Then the accuracy of forecasting 4 algorithms in both cases (in total, 8 predictions for return and 8 predictions for risk) compares and chooses the best algorithm. For this purpose data of 107 companies listed in Tehran stock exchange is used during the period of 2002 to 2014. The results show that in the case of using 16 independent variables, the linear discriminant analysis algorithm provides the best prediction for return and the quadratic discriminant analysis algorithm provides the best prediction for systematic risk. But in the case of using independent variables that are chosen, the quadratic discriminant analysis algorithm offers the prediction for return and linear discriminant analysis algorithm offers the best prediction for systematic risk. In general, using of selected independent variables (instead of using whole independent variables), improves the algorithm's ability in prediction of return and systematic risk.