The optimal threshold is the decisive factor in CAD mesh segmentation. It is difficult for a layman to set the optimal threshold for hybrid mesh segmentation. In this research work, a generalized technique is developed to predict optimal threshold (Area Deviation Factor) for CAD mesh model which makes hybrid mesh segmentation automatic by using a nonparametric, supervised learning classifier, i.e., K Nearest Neighbor (KNN). The proposed approach classify a CAD mesh model based on mesh attributes and predict the threshold. We demonstrate and validate the algorithm's ability to predict threshold using extensive testing on test models taken from various benchmarks, and it is found to be robust and consistent. We use percentage coverage and the number of primitives as a measure to test the efficacy of the algorithm. The experimentation shows that the optimal value of the threshold for segmentation results in better coverage. The KNN classifier predicts the threshold correctly with coverage of more than 95%. The novelty of the proposed method lies in threshold prediction based on mesh quality. The predicted threshold can be linked to a downstream application like automatic feature recognition from CAD mesh model.