Modelling claim severity and house price: an enhanced variable selection method

Teh, Jing Xuan (2024) Modelling claim severity and house price: an enhanced variable selection method. Masters thesis, Sunway University.

Preview

Text FInal Hardbound Thesis.pdf - Submitted Version Download (2MB) | Preview

Abstract

Variable selection is a crucial step in the model-building process in order to construct a suitable forecasting model. Variable selection techniques allow researchers to examine on the importance of every independent variable and generate the best subset of variables for the ultimate predictive model. Previous studies have shown that the filter and wrapper methods usually require greater computational resources. To the author’s best knowledge, the combination of random forest and least shrinkage and selection operator (LASSO)has not been investigated in the actuarial and financial industries, when evaluating the important factors for claims severity and house price. In this study, claims severity and house price data sets are used to build an enhanced variable selection method which combines random forest and LASSO approach. This study also compares various existing variable selection techniques using four data sets. The outcome demonstrates that the combined method (RF-LASSO method) yields superior results. The RF-LASSO method selects lesser independent variables to be integrated into the final forecasting model compared to other methods. Although the independent variables are lessened in the final model, however the R square value is not impacted much. The findings gained from this study might be of assistance to data analysts in insurance and finance industry who are interested in implementing variable selection. The findings are subject to several limitations such as not using discrete dependent variable. Future research should certainly further test whether the method of combining variable selection methods is effective for logistic regression.

Actions (login required)

View Item