Main Article Content
For development of software, controlling the quality is a prime concern of developers. Machine learning (ML) techniques enablesoftware engineers tocarry research in defect prediction that relies primarily on hand-crafted attributes, which are used to classify defect code in ML classifications. In the present studyML approach, a Recurrent Neural Network (RNN) for predicting software defect were used on PROMISE dataset for five different version of software defects.The proposed approach is compared with two existing approaches i.e. RF & SVM to predict software faults on historical data. Experiments on various Pledge datasets from the PROMISE repository: JM1, KC1, KC2, PCI and CM1 variants are studied. The proposed approach is better than the two existing support vector machine and random forest approaches in analysis. The RNN model in the present study shows accuracy in the range of 93.74 to 95.9 %. It shows maximum accuracy in PC1, and least in JM1. Further, the proposed approach gives precision in the range of 91.39 to 95.28 %. It shows highest precision in CM1, and least in PC1. Similarly, Recall is observed in the range of 92.21 to 94.64 % and thehighest recall is observed in JM1, and least in KC2.