Attribute Reduction Based on Fractal Dimension and Joint Entropy

Attribute reduction is a hot research problem in data mining, but classic attribute reduction algorithms and their extension methods all rely on the decision attribute, which can not be applied to attribute reduction of data set without label. In this paper, a combined attribute reduction method based on fractal dimension and joint entropy is proposed. To a certain extent, this method can solve the attribute reduction problem of a none-label data set. Experimental analysis shows that the method gives satisfactory performances on most data sets, can be used as a preprocessing approach for dimensionality reduction on high-dimensional data, or, a feature selection method on general data set.