An enhanced software defect prediction model with multiple metrics and learners

Abstract

Defect prediction is a critical technique for achieving high reliability software. Defect prediction models based on software metrics are able to predict which modules are fault-prone, which in turn. The prediction results would make the software developers to pay more attentions to these high-risk modules. For software defect prediction modelling, machine learning techniques have been widely employed. Model selection problem is always a challenge for generating an efficient predictor with a satisfied performance which is also always difficult to achieve. In this paper, a software defect prediction modelling framework based on multi-metric space and multi-type learning models is proposed. Different types of component classifiers and different software metric sets are used to build a software defect prediction ensemble model with the increment on the diversity of ensemble learning as far as possible. The proposed model is fully investigated by using a set of real project data from NASA MDP, the experimental results reveal that the model effectively improve the generalisation performance and the predictive accuracy.