Cost-Sensitive Boosting Networks for Data Defect Prediction

Software Defect prediction which classify the software modules into defect prone and not defect prone category and plays an important role in reducing the coast of software development and maintaining high quality software system. Existing defect prediction model face two challenges 1) Class Imbalance 2) High Dimensionality. In this paper a new cost sensitive boosting networks for data defect prediction is proposed. Cost sensitive learning is a method to solve class imbalance problem. Cost sensitive learning takes costs such as the misclassification coast into consideration. High dimensionality can be handled by using feature selection methods. For feature selection purpose three cost sensitive feature selection algorithms are used namely Cost-Sensitive Variance Score (CSVS), Cost-Sensitive Laplacian Score (CSLS), and Cost-Sensitive Constraint Score. The proposed techniques are evaluated on the data set taken from NASA MDP data set. The experiments shows that cost sensitive feature selection methods are more efficient than traditional one in reducing the total cost. The accuracy and class imbalance problem can be better solved by using the method like swarm and bagging techniques.