Summary Of Software Prefects

1744 Words7 Pages
Abstract Early predictions of software defects during software development life cycle play a vital role to produce successful software project. Besides, it reduces the cost of software development, maintains high quality, improves software testing and facilitates procedures to identify defect-prone of software packages. We need more research studies today to improve convergence across research studies due to its use a different of inconsistent measurements. Therefore, to avoid several types of bias: different sizes of the dataset are used in our study to avoid the bias of data. Comparative studies relied on appropriate accuracy indicators suitable for the software defect model are utilized to overcome the problems of bias error in…show more content…
Therefore, if we can identify which modules in software most probably to be defective it can help us to reduce our limited resources and time of development [2]. Based on the specific types of software metrics in the previous research studies, numbers of predictive classification model are proposed for software quality it can explore the defectives in some of the software modules by using several types of classifiers such as decision tree [3], SVM [4], ANN [5] and Naïve Bayes [6]. The classification model is one of the most common approaches used for early prediction of software defects [7], including two categories: Fault-Prone (FP) software and Non-Fault-Prone (NFP) software. The prediction model represented by a set of code attribute or software metrics [7, 8]. The objective of this research is utilized ensemble learning method that combines the vote of multiple learner classifiers by using the different subset of feature to improve the accuracy of supervised learning. Another advantage of ensemble methods, it enhances the performance by using different types of classifiers together because it reduces the variance between them and keeps bias error rate without increasing. Two main types of ensemble methods [9]: Bagging and Boosting techniques [10]. The first one depends on subsampling of the training dataset by replacement sample and generates training subset then

More about Summary Of Software Prefects

Get Access