A model to assess the effectiveness of fault prediction techniques for quality assurance

Fault prediction techniques aim to predict faulty module in order to reduce the effort to be applied in later phase of software development. Majority of the approaches available in literature for fault prediction are based on regression analysis and neural network techniques. It is observed that numerous software metrics are also being used as input for fault prediction. In this paper, a cost evaluation model has been proposed for Object-Oriented software which performs cost based analysis for misclassification of faults. Appropriately, this work focuses on inspecting the usability of fault prediction. Chidamber and Kemerer (CK) metrics suite has been considered to provide requisite input data to design the model using logistic regression and hybrid approach of Neural network and Particle Swarm Optimization (Neuro-PSO and Modified Neuro-PSO). Here, fault considered as dependent variable and CK metric suite are as independent variables. A case study of Eclipse JDT core has been considered for predicting a comparative study of performances of two approaches. Fault prediction is found to be useful where normalized estimated fault removal cost (NEcost) was less than certain threshold value. Modified Neuro-PSO model obtained promising results in terms of cost analysis when compared with those of Neuro-PSO and logistic regression.