BUILDING A CREDIT SCORING MODEL BASED ON THE TYPE OF TARGET VARIABLE

With the rapid development of big data and Internet technologies, companies engaged in financial platforms collect systematize massive through their own platforms, improve credit scoring parameters use machine learning methods to conduct complex scientific assessments. Thus, banks face problems when building scoring. Based on limitations existing system personal rating, it is necessary study rating based methods, clarify collection channels, dynamic desensitization technology. To reduce sensitivity data, LOF test method used verify emission random forest fill missing values. Then you gradient-boosting decision tree view important indicators, process proven indicators using a metric model logistic regression, get score. Finally, tested BP neural network, predict level credit. The shows that can further accuracy individuals' ratings provide basis background information for commercial banks' ratings.