Statistical Data Analysis and Machine Learning Methods in Insurance Fraud Detection Tasks

Over the last decade there has been an increased interest in the insurance industry in detecting different types of insurance fraud. Today, companies prefer to use employee experience as the primary method of fraud detection, and only 10% of companies surveyed use a very important tool such as text mining. This is why the aim of this paper is to develop a high-quality anti-fraud model. In particular, the model relies not only on conventional data, but also on textual information using deep learning techniques. In this paper, the raw data is processed, relevant features are selected, heterogeneity of data is checked using clustering method, neural network is constructed and its output is used as a variable, in addition, how textual data implementation affects predictive ability is checked and quality of old and new machine learning methods is compared. The results of this study have the potential to help reduce fraudulent insurance schemes, thereby reducing the overall risks of insurance companies, which may contribute to lower prices of insurance services and consequently increase insurance portfolios, which will reduce the cost of insurance by greater diversification of risks of companies.