Abstract
Abstract: Software defect prediction is considered as most interesting area for researchers in the field of software engineering. This process of defect prediction, identifies the bug during automated testing process which prevents the development of faulty software module. According to this process, previous archives of software modules are considered for analyzing the quality based on data mining and machine leaning concepts, which identifies the faults in software modules. Several techniques have been presented for software defect prediction using data mining techniques and machine learning techniques but achieving desired accuracy in performance is a challenging task for researchers. To address this issue, in this work we have presented a new approach for software defect prediction by combining genetic algorithm optimization process for feature subspace reduction and deep belief network for pattern learning. Deep belief networks are further improved by applying L1-regularization scheme resulting in better learning process by reducing the overfitting error. This combined model is implemented on SPIE lab software defect database. An extensive experimental study is carried out which shows that proposed approach achieves higher accuracy when compared with state-of-the-art software defect prediction techniques.
Highlights
Modern software modules and development techniques are growing rapidly
In order to deal with this issue, researchers have found that early prediction of bugs or code defects can improve the quality of the software application
LITERATURE REVIEW we review the studies based on the software defect prediction techniques
Summary
Modern software modules and development techniques are growing rapidly. Software based applications are widely adopted in real-time & daily life scenarios. Software defect prediction techniques utilize the historical data for further software bug prediction but due to insufficient dataset the desired performance cannot be achieved This issue can be addressed using metric set based evaluation for training and testing dataset. Lessmann et al [12] discussed about this issue and presented an experimental study using 22 classifiers which are tested for 10 publicly available databases This technique of software defect prediction depends on the feature extraction techniques. Software datasets suffer from class imbalance problem where learning becomes a very complex task This issue is addressed by using multiple kernel learning algorithm which helps to map the historical defect data into a higher dimensional feature space and express better and later ensemble learning can be implemented which uses series of weak classifiers to reduce the majority class and helps to achieve better prediction performance. As discussed before, learning becomes a crucial step where data distribution is random and imbalanced in nature
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
