A Novel Ant Colony Based Genetic Approach Neural Learning for Defect Prediction in Software Products

Software defects prediction is a process that involves using various techniques and models to anticipate and estimate the occurrence of defects or bugs in software during its development or maintenance lifecycle. The main goal of software defects prediction is to identify areas of the software that are more likely to have defects, so developers and testers can focus their efforts on those specific parts to improve the software's overall quality. By predicting potential defects early in the software development lifecycle, developers and testers can focus on addressing these issues before they become critical problems. This leads to reduced rework and lowers the cost of fixing defects. Predicting defect-prone areas helps in allocating testing and review resources more effectively. Instead of evenly distributing efforts, teams can concentrate on the parts of the software that are more likely to have defects, leading to better testing coverage. In most of the existing approaches, the prediction of defects are carried out using statistical models and machine learning models. Imbalanced datasets, where the number of defect-free instances significantly outweighs the number of defective instances, can lead to biased predictions. Statistical models may struggle to capture the patterns of defect-prone instances adequately, leading to lower recall or sensitivity. The proposed model implements genetic neural network for prediction of defects at early stages. ANNs are capable of capturing complex non-linear relationships between software metrics and defect occurrences. This is particularly advantageous when dealing with software defects, as the relationships between metrics and defects may not always be linear. Genetic algorithms can be used to automatically search for the optimal subset of features and fine-tune hyper parameters of the ANN. This helps in finding the most relevant features and achieving better predictive performance (Abstract)