A golden eagle-based hybrid deep learning model for automobile insurance fraud detection

Insurance fraud detection is a significant problem in the insurance industry, producing immeasurable losses. Conventional insurance fraud detection models depend heavily on experts' knowledge, and accurately estimating fraud when the data and the claim data are enormous is a complex and difficult task. This study proposes an efficient and effective Automobile Insurance Claim Fraud Detection (AICFD) approach. The feature selection process in the proposed approach uses Golden Eagle-Assisted Optimisation (GEAO) to efficiently select the subset of features. The obtained features are utilised for fraud detection using the deep learning model of hybrid Bidirectional Encoder Representation-Long Short-Term Memory (BERT-LSTM). The experimental analysis using the carclaim.txt dataset achieved better accuracy and recall of 99.02% and 99.1%.