Prediction of Insurance Claim Severity Loss Using Regression Models

摘要

The objective of this work is to predict the severity loss value of an insurance claim using machine learning regression techniques. The high dimensional data used for this research work is obtained from Allstate insurance company which consists of 116 categorical and 14 continuous predictor variables. We implemented Linear regression, Random forest regression (RFR), Support vector regression (SVR) and Feed forward neural network (FFNN) for this problem. The performance and accuracy of the models are compared using mean squared error (MSE) value and coefficient of determination (Rsquare) value. We predicted the claim severity loss value with a MSE value of 0.390 and a Rsquare value 0.562 using bagged RFR model. In addition where applicable, the final loss value was also predicted with an error of 0.440 using FFNN regression model. We also demonstrate the use of lasso regularization to avoid over-fitting for some of the regression models.