Artificial Intelligence-Based Fraud Detection Technique Using Unsupervised Machine learning for Real Estate Property Insurance

Abstract

When it comes to money, insurance fraud is a serious issue and still a challenging area of interest. This study proposes that Artificial Intelligence (AI) and its branch of unsupervised machine learning can effectively detect such fraud. A dataset of 1,070,994 property records with 32 structured data fields, location, value, and relevant features, was chosen using clustering algorithms. In addition, Principal Component Analysis (PCA) and an autoencoder for retaining essential information were technically used in this study. The top ten properties exhibiting the highest fraud scores were identified. This study includes other factors to identify anomaly scores for unusual patterns. They are extremely low or high values. The results were significantly high, at almost 97%. This system opens up a direction for researchers to identify ambiguities or fraudulent claims in insurance systems.