Insurance fraud represents a source of significant financial loss for insurance companies, particularly auto insurance fraud. Indeed, the insured seeks to increase his indemnity by fraudulent acts or declares a false claim. Consequently, the detection of automobile insurance fraud holds an important place in the strategy of the insurance company, since it allows to reduce the costs of claims and maintain a satisfactory profit. With this in mind, insurers are always looking to have more efficient detection systems in order to overcome the limits of traditional methods.

In this paper, we propose an automobile insurance fraud detection system based on Machine Learning. Two unsupervised learning algorithms will be used, namely Isolation Forest and Local Outlier Factor.

According to previous studies, these two algorithms are not yet applied in auto insurance fraud detection. Both algorithms belong to the anomaly detection algorithms family which is more suitable for this type of problem.

The results of this study show the possibility of developing a detection system, based on these two Machine Learning techniques, which serves to automatically detect fraud with considerable accuracy.