Abstract:
Software defect prediction is significant contributor in better understanding and controlling
software quality. Many organizations want to predict the number of defects in software systems
before they are deployed, in order to gauge the likely delivered quality and maintenance effort. A
defect prediction model’s main objective is to identify error-prone parts of a software system as
early during software lifecycle as possible, since costs of finding and correcting software defects
later in the lifecycle rise prohibitively. Accurate prediction of defect‐ prone software modules
can reduce costs and improve software testing process by focusing on fault- prone modules.
Ant Colony Optimization (ACO) algorithm is a meta-heuristic technique for finding optimal
search paths in graphs. This algorithm is a member of swarm intelligence methods, and mimics
the behavior of ants trying to find a path between their colony and a food source. It is a very
good combinatorial optimization method, possessing attributes like parallelism, convergence to
good solutions, , and ease of hybridization with other methods.
In this research, ACO-MB a combination of ACO along with modified bagging technique is
proposed for software defect prediction. The proposed approach first assesses improvement in
training through bagging hit and trial approach and then only in case of improvement it trains the
predictor using ACO and records misclassification rate. This process continues as long as
improvement in the predictor performance is found.
