To reduce the cost of manual labor in various industries, machine-learning algorithms are widely used. The expected frequency of claims and the expected severity of consumer claims are used in predictive modeling for car insurance claims. Insurance companies use express insurance and loss settlement, where customers can independently upload photos taken on mobile devices. This type of insurance compensation is considered as a claim and can be processed both manually and automatically. Due to the daily increase in the number of claims, the processing of data is complicated; the processing of repeated claims regarding the case identity is complicated, which can lead to large losses for insurance companies.

Two types of claims category are considered:

  • Damage to third party property (TPPD)
  • own damage (OD).

Currently, neural networks have a high ability to receive non-linear, complex and stochastic data. Neural networks show high accuracy compared to traditional methods and are preferred in data mining.

In the field of car insurance, fraud is very common. Fraud comes in many forms and sizes, from traditional fraud, such as (simple) tax fraud, to more sophisticated fraud, where entire groups of people collaborate to commit fraud. Fraudsters arrange road accidents and issue fake insurance claims to get funds from their General or car insurance. There are cases where vehicles are placed only on the road, initiating a traffic accident. However, most such fraud is not planned – the person only takes advantage of the opportunity arising from the accident and issues exaggerated insurance claims or claims for past damage.

Insurance companies are most interested in organized groups of fraudsters, consisting of drivers, garage mechanics, lawyers, police, insurance workers, and others. Such groups represent a large part of the revenue drain. Conducting an analysis of claims revealed that about 20% of all insurance claims to some extent fraudulent. However, most claims go unnoticed, as fraud investigations are usually conducted manually by an expert or investigator. Thus, an expert system approach is needed.

Manual processing for large-scale claims is time-consuming and does not meet the speed requirements for the Express claims process and automatic reporting of duplicate candidate claims before settlement is required. However, this solution remains a challenge due to a large number of factors: in video frames captured by mobile devices, the point of view and lighting conditions are not taken into account in the controlled environment. Automatic or partially automatic processing of insurance claims can be very useful in the insurance industry when considering small but more frequent insurance claims under a certain scheme.

Data analysis experiments are performed by developing predictive models. Forecasting is a statement about how things will happen in the future, often but not always, based on experience or knowledge. Predictive modeling includes problem identification, data analysis, model development, and forecasting. Predictive modeling in the insurance industry helps insurance analysis by using predictive models to improve business operations that previously used human experience.

Artificial neural networks are used with a high level of success in forecasting. Artificial neural networks have a natural cumulative function and make knowledge available for further use in forecasting. They have the ability to approximate almost any nonlinear and complex function with any required accuracy. They search databases for hidden patterns, finding predictive information that might be missed by experts because it doesn’t meet their expectations. Various types of neural networks can be used for prediction, such as multi-level perceptron, ART, radial basis functional networks, and others.