Severity of defect: an optimised prediction

Abstract

To assure the quality of software an important activity is performed namely software defect prediction (SDP). Historical databases are used to detect software defects using different machine learning techniques. Conversely, there are disadvantages like testing becomes expensive, poor quality and so the product is unreliable for use. This paper classifies the severity of defects by using a method based on optimised neural network (NN). In full search space, a solution is found by many meta-heuristic optimisations and global search ability has been used. Hence, high-quality solutions are finding within a reasonable period of time. SDP performance is improved by the combination of meta-heuristic optimisation methods. For class imbalance problem, meta-heuristic optimisation methods such as genetic algorithm (GA) and shuffled frog leaping algorithm (SFLA) are applied. The above method is based on SFLA and the experimental outputs show that it can do better than Leven berg Marquardt based NN system (LM-NN).