A personal credit forecasting method based on improved isolation random forest
Rongxin Guo, Sihao Fu, Peilong Guan · 2021 IEEE 3rd International Conference on Frontiers Technology of Information and Computer (ICFTIC) · 2021
In recent years, due to the development of financial business and the Internet, personal credit business has become the focus of financial enterprises and the public. With the development of artificial intelligence technology, data-driven methods have been widely used in personal credit forecasting. However, due to potentially incorrect personal information about consumers, or possible errors in the collection of data, the credit system has some abnormal data in its database, which may lead to a decrease in the accuracy of the forecast. Furthermore, traditional machine learning methods ignore the influence of outliers on the model. In this paper, an improved isolation random forest model is proposed for personal credit forecasting. First, an isolation forest is used to isolate anomalous data to reduce the misguidance of the model. Compared with the general anomaly detection algorithm, the isolated forest does not need to make assumptions about the distribution of data and can effectively detect anomalies in the data. Then, the random forest is applied to build personal credit models based on the clean data. Compared with other methods, finally the accuracy of the proposed method is at least 6.57% higher, even under the interference of strong noise.