Multi-modal data-oriented car insurance fraud behavior prediction system, method and device

Abstract

The invention discloses a multi-modal data-oriented vehicle insurance fraud behavior prediction system, method and device, and the method comprises the steps: extracting risk factors from picture data, combining the risk factors with corresponding structured data fields, constructing a vehicle insurance fraud risk prediction model based on the algorithms of feature engineering, machine learning, deep learning and the like, and predicting the fraud behavior of the vehicle insurance. And carrying out early warning on a risk behavior. After prediction, risk assessment and importance ranking are carried out on the picture factors, and visual expression is carried out on the factors with high risks and high weights. According to the method, manual risk assessment can be effectively assisted, and visual causal relationship expression of the model and the prediction result is realized by using data of different types of pictures. According to the method, a computer vision algorithm is utilized to perform factor extraction on some picture data difficult to use, and a prediction model and a result are visually displayed by means of factors analysis, causal inference and other algorithms.

Description

technical field [0001] The present invention relates to the field of identifying insurance fraud, in particular to a multi-modal data-oriented auto insurance fraud prediction system, method and device. Background technique [0002] As the frequency and losses of insurance fraud cases are increasing year by year, the situation of insurance fraud is becoming increasingly severe, and the detection of insurance fraud is of great significance. Therefore, it is of great significance to re-identify image information such as investigators, drivers, and auto repair shops. [0003] Most of the existing computer identification technology applications in the vehicle insurance industry are aimed at insurance loss determination and on-site records, and there are few applications that directly extract risk factors for vehicle insurance fraud based on various picture information. The vast majority of fraudulent judgments based on photos are done manually. For the pictures, texts and other...

AI Technical Summary

Problems solved by technology

For the existing machine learning and deep learning models that predict whether a case is a fraud case, they do not comprehensively use multi-modal data (text, documents, photos, etc.) for modeling, resulting in inefficient use and loss of key information. And generally not interpretable, front-line practitioners tend to tend to be more conservative in judging the results of model training

Images