An Adaptive Optimization Method of Configuring Feature Weight Group

It introduces a blended objective function that can represent the guideline more precisely, data points in any cluster are close to each other, and data points between any two clusters are away from each other. By optimizing the blended objective function to get an optimal feature weight group, we can construct a classifier based on clustering with an optimal distance measure. In order to seek an acceptable solution of the blended objective function, it gives an adaptive optimization method of configuring feature weight group based on reverse projection of grads of the blended objective function. The method is valid by experiments of two data sets from UCI. At last, it gives some analysis and discussions of the method and points out that it can be applied to continuous attributes reduction