Implementation of a Predictive Model for Fraud Detection in Motor Insurance using Gradient Boosting Method and Validation with Actuarial Models
Rohan Yashraj Gupta, Satya Sai Mudigonda, Phani Krishna Kandala, Pallav Kumar Baruah · 2019
Machine Learning provides greater ability to identify in-depth patterns in the data that are normally invisible or difficult to identify using other methods. One of the major application of it is seen in insurance claims fraud detection, which is a classification problem. In this work, Gradient Boosting Method (GBM) was used to create a predictive model which was applied to motor insurance claims data. The dataset was highly imbalanced; this problem was addressed using Synthetic Minority Oversampling Technique (SMOTE). The results achieved were remarkable with F1 score around 98% and the accuracy 99%. This was cross-validated by industry experts using extreme value theory (EVT), an actuarial model. The predictive model presented in this paper can be customized, tested and extended to other lines of business.