Supervised machine learning-based security financing account identification method

Abstract

The invention discloses a supervised machine learning-based security financing account identification method. The method comprises the steps of a), acquiring and marking a sample; b), extracting and calculating transaction features; c), dividing a training set and a test set; d), performing feature screening and validity sorting; e), performing model training; f), performing model verification; g), increasing a feature number in a model and performing re-training; and h), changing a classification method or a classification model, and determining an optimal classification model. A financing account type judgment process comprises the steps of 1), calculating all the transaction features contained in a feature set MY of a to-be-predicted security account; and 2), inputting the features calculated in the step 1) to the optimal classification model Y, thereby obtaining a financing status of the to-be-predicted account. According to the security financing account identification method, the manpower input required for investigating and dealing with illegal financing is reduced, about 60% of financing accounts can be identified, the accuracy is about 80%, and the workload of manual investigation and evidence collection is reduced; and therefore, the method is of important practical significance.

Description

technical field [0001] The present invention relates to a securities allocation account identification method based on supervised machine learning, more specifically, it relates to a method of using marked allocation account data to train a base classifier, and by analyzing the allocation account of the verification set Features are predicted, the optimal base classifier is screened, and the method of discriminating whether there is an illegal allocation of funds in the securities account is realized. Background technique [0002] Since 2015, the China Securities Regulatory Commission has carried out a strong clean-up campaign against off-market fund allocation activities, and has severely investigated and dealt with umbrella trust software represented by Macktronic, Flush, and Homs Software. On July 12, 2015, the "Opinions on Cleaning up and Rectifying Illegal Securities Business Activities" issued by the China Securities Regulatory Commission clearly stipulates that "accou...

AI Technical Summary

Problems solved by technology

There are three shortcomings in this identification method: first, the screening criteria for the allocation of capital accounts is relatively simple, generally using indicators such as average value and variance, and there is no good integration method between multiple screening conditions; second, it is subjective and generally relies on The business understanding of supervisors is used to screen, and there is no uniform standard for screening and marking, which makes it difficult to promote; third, it requires high manpower input. Due to the low accuracy of the first step of statistical screening, a large amount of screening work is actually left to manual investigation. stage

Images