Algorithm research about feature selection based on fractal dimension

The tradition backward feature selection algorithm based on fractal dimension need to calculate the fractal dimension of various projection sets about the datasets and to scan dataset many times that make the algorithm low efficiency and high complexity. By analyzed the nature of the individual contribution that an attribute affects the intrinsic dimension of a dataset, a forward feature selection algorithm and a two-way feature selection algorithm were proposed based on fractal dimension. The two algorithms realize the feature selection all based on the concept of attribute individual contribution to the intrinsic dimension of a dataset. The experiment results show that the forward feature selection algorithm and the two-way feature selection algorithm proposed in this paper can reduce the times of calculating the fractal dimension a lot than the backward feature selection algorithm and lower the algorithm complexity effectively. The lower space complexity algorithm to calculate the fractal dimension was proposed too in this paper.