This work presented a defect classification methods based on improved classification algorithm in additive manufacturing process. To make the algorithm be applicable in process monitoring tasks, a method of optimizing the evolution process in GP evolution was raised in this work. A series of specific functions and their linear combinations were introduced to represent the GP classification model. The evolution process in this strategy is designed to optimize the coefficients of these functions and the offset. The advantaged in GP are also completely inherited. Comparing with GP alone, the improved strategy could reach higher classification accuracy in engineering application, i.e., process monitoring of additive manufacture.