Analysis of Bio Inspired Based Hybrid Learning Model for Software Defect Prediction

Analysis of Bio Inspired Based Hybrid Learning Model for Software Defect Prediction

연구 분야: Verification

학회: SN Computer Science

The software’s quality can be ensured through software testing, which is one of the critical methods. However, it was found that testing consumes more than half of the project's total expenses. Effective and efficient software testing utilizes minimal software resources to find as many flaws in the software system as possible. This paper presents research on software defect prediction using hybrid machine learning algorithm. The algorithm incorporates the benefits of supervised as well as unsupervised learning methodology that improves the accuracy of defect prediction. The study uses publicly available datasets from the open source PROMISE repository provided by NASA MDP to perform machine learning analysis and maximize accuracy. In particular, the study suggested a hybrid learning model and conducted a comparative analysis among Naïve Bayes, Multilayer Perceptron, Random Forest, CNN, Adaboost, and KNN on four different datasets namely CM1, KC1, KC2 and PC1. With respect to F1-score, accuracy, precision, and recall, the results show that the hybrid method performed better than the other methods. The study achieved an accuracy of 93.67% (BAT + CNN) on PC1 dataset, 92% on (Bat + CNN) for CM1 and KC1 respectively. This indicates that the hybrid learning technique of (BAT + CNN) is more potent than other methodologies in prediction of defects. The research can help software development teams identify potential bugs early on, leading to efficient and effective software creation. One useful approach for predicting software defects is the hybrid machine learning method, which can lead to improved software quality and lower maintenance costs. This aids the software testing phase by predicting the faulty modules with better accuracy thereby reducing the testing effort. The main objectives of the research include applying different machine learning algorithms and methods to the given datasets and determine the most effective ones.