In this paper, Wisconsin breast cancer dataset is taken from UCI to minimize its features. It has thirty input variables and one output variable. In earlier, the prediction of breast cancer is made by machine learning algorithms like linear regression, neural network, decision tree, SVM and so on. Here, the features or input variables are reduced to eleven input features from thirty-two through similarity measure and optimization method. For this, first Pearson correlation is applied between the variables and the attributes are reduced when its pair has a 90% correlation. Then, Cost Optimization based Machine Learning algorithm is applied to the constraint pairs. From this result, it has observed that we can predict breast cancer with only two input features. The error rate and accuracy of various classifiers are also presented here. © BEIESP.