Abstract
Fraud can take on various forms, including financial fraud, identity theft, and insurance fraud, among others. With the growing use of technology, fraudulent activities have become more sophisticated, making it difficult for organizations to detect and prevent them. One major challenge in the insurance industry is vehicle insurance fraud, which leads to increased expenses and a loss of trust. Using machine learning techniques has gained prominence as an efficient approach for detecting fraud. This paper aims to test the performance of various supervised machine learning models using different data resampling techniques (undersampling and oversampling) for vehicle insurance fraud detection. This study compares the performance of NearMiss, SMOTE, and our proposal hybrid data augmentation approach for data resampling. The preprocessing steps used in the methodology include dropping irrelevant features, filling missing values, encoding features with dummy variables, and selecting features using a correlation approach. The testing results indicated that of Random Forest (RF) model performed best using our proposal hybrid data augmentation approach achieving the highest F1-score of 0.975 and accuracy of 0.975 in fraud detection.
Boulbaba Ben Ammar
Assistant Professor
Boulbaba Ben Ammar is an Assistant Professor at the Computer Science Department, University of Sfax, Tunisia.
Mohamed Ben Aouicha
Associate professor
My research interests concern information retrieval, semantic technologies, social media analytics, knowledge representation, Big Data and graph embedding.