Interpreting Cabinet Approval Ratings by Neural Networks

R. Kamimura, F. Yoshida, and R. Kitajima (Japan)


Approval ratings, early stopping, information loss, rele vance, BP, regression


In this paper, we try to estimate Japan’s cabinet approval ratings by using neural networks. In addition, we try to extract the important features in input patterns. This is the first attempt to use neural networks and to interpret the mechanism of inference for approval estimation in a com prehensive way. Experimental results show that neural net works have much better performance than that obtained by the standard regression analysis in terms of training and testing errors. The information loss analysis reveals that the first variable, that is, the previous ratings should play the most important role in inference. Though the experi mental result here shown is a preliminary one, it certainly suggests a possibility of the automatic inference of cabinet approval ratings.

Important Links:

Go Back