A Comparative Analysis of Three Supervised Learning Algorithms in Stock Selection

In this paper, our goal is to judge which algorithm is the best, through comparing the classification accuracy of the three supervised machine learning algorithms, using the data of four financial factors which can reflect the intrinsic value of corporate stock. Our empirical results show that Support Vector Machine got the extremely high classification accuracy in the test both inside and outside the samples; Random Forest achieved the highest classification accuracy in the test within the samples, but it's accuracy was not as good as Support Vector Machine in the test outside the samples, it means Random Forest was prone to over-fitting; The classification accuracy of Naive Bayes was very low in the test both inside and outside the samples. According to the Bayesian hypothesis, we can see that financial factors are not independent of each other. As a result, the optimal algorithm is Support Vector Machine, followed by the Random Forest, and it is not advisable to use the Naive Bayes, when we selecting the stocks using financial factors data.