Subtraction & Rotation Algorithm And Application For Multi Factor Classification Problems

For the classification problems on uncompleted factor sets,to improve the effectiveness of classification algorithms,a new multi-factor algorithm named S&R is proposed.It follows the basic principle of hypercube cover of nonlinear classification,puts the significance of set include in data decision and set symmetric in information identification as guideness,creates the branch rules of decision tree by observing the division ability of condition factor to decision area.The description and steps of S&R in single factor condition(dispersed or continuous)are presented.The feasibility is analyzed in several UCI data sets,and the results are compared with C4.5 and BP neural network.The results show that S&R is basically successful.For error rates,S&R is equal to decision tree C4.5,and much better than BP neural network in limited learning times.For time complex,because it is on different comparing level with C4.5,the results need research more.But it is absolutely better that BP in effectiveness and stability.S&R is applied in factor reduction problems.The reduction results of S&R are compared with identification matrix based rough set reduction method in several discrete UCI data sets.It shows that S&R has a good factor reduction effectiveness.