Does a Back Propagation Network and Grey Clustering The Right Combination to Forecast Policyholder Decision?

By Fakta Liza

Student of S1 Statistika

Universitas Muhammadiyah Semarang (UNIMUS)

The policyholder of investment-linked insurance bear the investment risk, their risk attitude should have a great impact on their purchase decision. Therefore, backpropagation network is apply to forecast policyholder decision of investment-linked insurance and compare the results with that of logistic regression.

First of all, i think that the performance of both the back propagation network and logistic regression depends on the method used in data transformation process. But, backpropagation more dominate the logistic regression so that the level of accuracy of the combination backpropagation network with gray clustering is higher than the standard deviation method with logistic regression. So, grey clustering is better suited to back propagation neural networks; while the average-and-standard-deviation method is better in combination with logistic regression.

In addition, research shows that the most widely-used methods for classifying risk attitudes are based on either the summations or the average and standard deviation. However, this method has many errors and gray clustering provides an alternative classification. The estimated mean of the accuracy ratios of model 1, composed of grey clustering statistic and a back propagation neural network, is 82.53%, the estimated mean of the accuracy ratio of model 4, composed of the average-and-standard-deviation method and logistic regression, is 82.28%.

To sum up, I believe that the gray clustering approach offers an alternative to existing classifiers of risk attitudes, while the neural network approach offers an alternative to existing purchase decision prediction models and back propagation network combined gray clustering to forecast policyholder decision to purchase investment-inked insurance can produce optimal performance.