Abstract: Existing feather and threshold selection lack quantitative models and methods in license recycling rule.We analyze feature sensitivity and correlation by such statistical methods as differential coding, information entropy, correlation matrix, based on formalization defined parameter model.We also define feathers and threshold with sensitivity,cost minimization,convergence consistency and stability principles.The method can select CPU utilization rate as the only feature item from several candidates and set the threshold of 8% and statuscheck interval of 1 136 s.Experimental results show that it is simple, effective and easily programed.

Key words: statistics, entropy, rule definition, feature selection, correlation coefficient