Proposed and tested an entropy impurity based objective function for incorporating a learnable perceptron into the decision tree framework. Demonstrated that the learned classifier achieves comparable accuracy with fewer test time computations than an MLP