Fraud Analysis Using Machine Learning Algorithms

Abstract

In the previous studies done by others, the methods of picking abnormal numbers and data have been developed. This paper will focus on using machine-learning method to find abnormal value. Machine learning has been applied in all kinds of fields, including picture identifying and regression which is widely used in finding anomaly. The author finds some of the anomalies during the investigation and the fraud detection rate is higher than 80 percent. This paper not only aims at using traditional way of finding abnormal value, but also aims at using the modern computer science to deal with huge amount of data. Traditionally, when it comes to the topic of finding anomalies, people tend to think about t-checking or f-checking[1,2], which shares the same limitation that the target has to be fixed when being measured. However, a more realistic problem is that the target and the measuring environment is always changing. To fix the problem, this paper uses machine learning to do regressions of the environment and the target. Not only can it help us find the relationship between our target and 'environment', but also it can be written into computer elegantly so that it could be able to deal with huge mount of data.