Feature Mining through Distance Minimization Learning

摘要

A common criticism of neural net learning models is their lack of explanation facilities such as are used in expert systems. Providing explanation tools for such systems would eliminate such criticisms, while also helping to facilitate learning more about the problems that these models solve. This paper establishes a model that provides classification via supervised and unsupervised learning and further supports mining information regarding the importance of the input features in training an optimal map. The features are ranked in importance and both simple and higher order correlations are displayed. Once trained, this model then returns the information on the roles that the features played in determining output classification in both supervised and unsupervised learning models.