To improve software reliability, software should be developed without defects. SDP models can be employed to identify defective code sections at initial stage during the software development. When defects are ascertained early, it helps the practitioners to prioritize the efforts for testing and allocating more resources to defective modules. This results in improved software quality, reliability, and efficiency. Despite the fact that the numbers of approaches have been used in the past for SDP but most of them are not practically applicable. Manual feature selection is mostly performed by majority of feature selection methods. Core aim of this research is to propose iterative feature selection technique using Boruta (random forest) for SDP model. This research proposes Two?step preprocessing using SMOTE and BORUTA. Support Vector Machine (SVM), Neural Networks and XGboostclassifiers are used by MLC. Furthermore to confirm the accuracy, performance and capability of each classifier on PROMISE dataset evaluationmeasures AUC,recall, F1-measure, andaccuracy are used