With an increase in the size and complexity of the software, software testing is a vital activity in software engineering to measure software quality. Finding and fixing defects in software modules have a significant impact on the cost of development and maintenance of the software product. Software defect prediction (SDP) is the process of finding defective components in software prior to deliver the software product to the customer. So that the quality assurance team can effectively allocate minimum resources for testing the product by setting more effort to the defective source code. In this regard, a wide range of Machine Learning (ML) models has been developed to predict defects in software. However, those SDP models have inadequate performances due to challenges like the presence of redundant, irrelevant features, and class imbalance problem. Class imbalanced occurs with data sample from two groups, the minority group contains considerably smaller samples than the majority group. The class imbalance nature of the defect data increases the learning difficulty of the classification algorithm to train the model.The use of imbalanced data leads to off-target predictions of the minority class, but which is considered to be more important than the majority class.Thesechallenges depreciate the performance of the defect prediction model depending on the predictor’s ability to tackle data frauds. In this study, we proposed a software defect prediction model that addresses class imbalance problem using Filter-Based Feature Selection (FBFS), Synthetic Minority Oversampling Techniques (SMOTE)and Support Vecctor Machine (SVM) algorithms. FBFS is used for selecting the relevant software features, SMOTE is used to produce balanced data.SVM is used for classification in which the use of Radial Basis Function (RBF) kernel function that enables the SVM classifier to maximize the optimal marginbetween the minority and the majority class.The main contribution of this study is application of FBFS and SMOTE sampling together that enables proposed model can effectively solve the binary classification faults from the minority and the majority class equally. To assess the performance of the proposed approach, we did experiments on six highly imbalanced datasets from a public NASA repository. The experimental resultsindicated that the proposed model makes an impressive improvement in SDP performancewhen compared with MAKAHAL, DNN Hybrid and EUS Adatptive related state-of art models, in which 99%, 99%, 100%, 99%, 99%, 99% accuracy is attained for KC1, MC1, JM1, PC1, PC3, PC5 datasets respectively. Thus, we conclude that the proposed model improve the performance of SDP effectively, and it provides a brand new way of dealing with the imbalanced data problem.