As the size of the defects increases, it becomes difficult to predict the different types of software defects with high true positive rate. The main objective of the machine learning models for software defect-based testing application is to improve the defect prediction rate with less error rate. Evaluating the software metrics and defect prediction are the two key quality features that determine the success of a software product. Most of the conventional meta-heuristic based software defect testing models are independent of dynamic parameters estimation. Also, these conventional models are used to predict the defect in the homogeneous software testing systems with limited number of feature space. In this paper, different types of software defect prediction systems and its models are discussed along with the limitations on various metrics.