The cost of fixing a software defect varies with the phase in which it is uncovered. Defect found during post-release phase costs much more than the defect that is uncovered in pre-release phase. Hence defect prediction models have been proposed to predict bugs in pre-release phase. For any prediction model, there are two kinds of misclassification errors - Type I and Type II errors. Type II errors are found to be more costly than Type I errors for defect prediction problem. However there have been only few studies that have considered misclassifications costs while building or evaluating defect predictions models. We have built classification models using three cost-sensitive boosting Neural Network methods, namely, CSBNN-TM, CSBNN-WU1 and CSBNN-WU2. We have compared the performance of these cost sensitive Neural Networks with the traditional machine learning algorithms like Logistic Regression, Naive Bayes, Random Forest, Bayesian Network, Neural Networks, k-Nearest Neighbors and Decision Tree. We have compared the performance of the resultant models using cost centric measure - Normalized Expected Cost of Misclassification (NECM). © Springer International Publishing AG 2018.