Classification of Stock Information Disclosure Level Based on SVM and CNN

Information disclosure is an important measure of stock market supervision, but the information characteristics of each listed company are varied, so how to classify and predict the information disclosure level of the company is a challenging task. In this paper, an information disclosure evaluation method of A-share listed companies based on One-Class Support Vector Machine (One-Class SVM) and Convolution Neural Network (CNN) models are proposed. The samples of D level companies (i.e., companies with unqualified information disclosure level) are insufficient, and they are the focus of supervision, so this paper adopts One-Class SVM to identify them from the other types of companies in the first place. Further, for other levels A, B and C (i.e., companies with qualified information disclosure level), the number of samples is relatively large, so the CNN classification method is adopted. By selecting nearly 2,000 stocks in the main board of A-shares as research objects, they are divided into training sets and test sets, and the model is trained according to various indicators of listed companies. Finally, the classification accuracy of the D category is about 81%, meeting the needs of market supervision.