Property and Casualty insurance companies often encounter problems in predicting the likelihood of a policyholder causing a claim. Some territories have few claim experiences, resulting in very sparse data. In addition, some data are highly dimensional in terms of the predictors of the likelihood of a claim.
In this research, simulated sparse claims data are used to identify the most appropriate predictive model for determining the likelihood of a policyholder causing a claim. Machine learning algorithms such as the logit model and the support vector machine will be used to predict whether the future policyholder will incur a claim.
http://prezi.com/2smclr8un8vu/?utm_campaign=share&utm_medium=copy&rc=ex0share
This is the main file of this project. The algorithm in this file requires some external R libraries, which are already mentioned in the comments in these files. The algorithm also gives us a choice to run either Support Vector Machine(SVM) or Logistic Regression(LR) and this can be done by commenting SVM or LR source file code, respectively.
This file is to process the generated data and split it into training and test data.
This file generates the data of specific number of observations and feature ratios.
The file runs the SVM classification on the training data and then test the classifier on the test data to check the accuracy of the classifier for the unknown data.
The file runs the LR classification on the training data and then test the classifier on the test data to check the accuracy of the classifier for the unknown data.