Application of Extreme Learning Machine in Detecting Auto Insurance Fraud

This paper deals with developing a novel methodology for detecting anomalous claims in auto insurance records by a neural network based Extreme Learning Machine (ELM). Initially, the raw dataset has undergone a preprocessing proce-dure and divided into the training, validation and testing sets. A pool of trained ELM classifiers is then generated by using different combinations of ELM parameters on the train set. The best ELM model is then selected by subjecting the validation set upon the trained models. Afterwards, the testing set is fed to the validated model for determining the nature of the insurance claims - legitimate/malicious. The performance of the model is demonstrated by extensive tests carried out on a widely used auto insurance dataset.