Many data mining applications, ranging from Spam filtering to intrusion detection, are fighting with active adversaries. Adversaries deliberately manipulate data in order to reduce the classifier’s accuracy. Consequently, in all these applications initially Successful classifiers will degrade easily. In this paper we model the interaction between the adversary and the classifier as a two person sequential Stackelberg game and analyse the payoff when there is a leader and a follower. Then we model the interaction as an optimization problem and solve it using evolutionary strategy. Moreover, to achieve better performance in Spam filtering, an adaptive cut-point value is used to separate Spam from Ham in an elegant manner. Adaptive cut-point is incorporated into the model by the mean of non-linear functions rather than that of linear ones. Non-linear functions offer our approach locating the optimal value that minimizes false negative rate (FNR) errors associated with binary outcomes. We investigate thoroughly the performance of our approach on real world Spam email datasets under different strategies. Our observations show that adaptive cut-point selection via non-linear functions brings significant benefits to Spam filtering in term of well known performance measures like accuracy and cost sensitive evaluation measures.