The Availability Research Of Generalized Regression Neural Network In Insurance Company’s Indication

The insurance industry has great improved in last decades in our country. Although we still have a large space to catch up with Europe and America. With the development of our economy and the improvement of peoples’ lives, the premium income has increased o lot. But the profit of insurance companies did not increase in the same speed. To realize which factors affect the profit of insurance companies, if these factors could indicate the profit, and how to indicate profit more precisely in Chinese insurance market-which is full of vitality but also behindhand- to improve the operating strategy, is a very important task.The source of China Life Insurance Company’s profit and its factors was studied in this paper. We used generalized regression neural network of extracting weight by the model named grey correlation analysis(Grey-GRNN) to predict the profit of the insurance company, and its prediction accuracy was compared with gray model, BP networks, GRNN networks and their combination weight prediction model(GM-GRNN,BP-GRNN,GM-BP-GRNN).Then the result was get as below. The Grey-GRNN network, which overcome problem of small samples, high dimension, multiple disturbance in profit prediction of insurance company was obtained better performance both in the historical data fitting and model extrapolation than other prediction models by reducing the dimension of original data via the model named euclidean distance optimization, weakening randomness of original data through gray treatment and setting the proper smoothness factors. This special network, with excellent ability on self-learning and re-recognition, was very useful in the prediction of insurance company’s profit to characterize the complex nonlinear relationship between profit and its factors. For Grey-GRNN network, the relative errors of training data fitting were all less than 5% in examples. From the result of extrapolation simulation,the MAPE value of neural network series model were all less than 10%,which belongs to the high precision prediction. The MAPE value of GM(1,1) reaches at 13.3042%(>10%),and it belongs to good prediction range. Compared with six neural network, there was almost no difference in MAPE value between the BP network, GRNN network and their combination weight prediction model(GM-GRNN,BP-GRNN,GM-BP-GRNN).However, the Grey-GRNN network can obtain an excellent result of MAPE value of 1.7638%,which represent the highest extrapolation accuracy in all models.The intension value, business cost rate and claim rate was the high weight factors calculated by the model of grey correlation analysis. The relationship between the high weight factors and profit of insurance company can be quantized by sensitivity analysis of neural network and qualitative analysis. The result reveals the internal connection between high weight factors and profits of insurance company.