Application Of Machine Learning In Credit Intelligence Assessment

The continuous improvement of the social credit system has made higher requirements for scientific and objective credit construction standards.In addition to the rapid expansion of data,credit evaluation can be further developed based on statistical methods and machine learning.Due to people’s daily demand for networks,it can be said that massive amounts of communication-related data are generated all the time.Therefore,this theses selects China Mobile’s customer data for an empirical analysis of credit evaluation.Firstly,this theses introduces the background and significance of credit evaluation construction,reviews the existing model methods of credit evaluation,and describes the machine learning method based on tree model.Then,combined with the actual business background and exploratory analysis of the data,we perform data preprocessing and feature engineering.Among them,missing value processing and new feature construction are the priorities to improve the prediction effect.Finally,based on five-fold cross validation,machine learning models including random forest,XGBoost and Light GBM are used for training to establish a credit evaluation prediction system.The features with weak correlation are randomly combined into multiple sub-feature data sets,and linear regression is used for stacking model on the previous training results.Based on the mean absolute error(MAE),stacking model performs better than the other models in terms of the prediction error of credit scores.In addition,the training results show that the characteristics such as length of time using the network,user age,the number of people in the conversation circle of the current month,relative consumption of the current month,relative stability,the proportion of online shopping applications are the important indicators of credit evaluation,which need extra attention in the future credit evaluation.This also shows that the new feature construction is meaningful.