APPLICATION STUDY OF ALGORITHM C4.5, MLP, AND NAIVE BAYES FOR SOFTWARE DEFECT PREDICTION

Abstract

Companies and institutions in various fields require software to help their business processes in order to run fast, precise, effective and efficient. Surely the software used must have good quality standards for the purpose of a company or institution can be met. Thus the necessary software that does not have a fault / error (defect). There have been many other researchers doing modeling for software development, which will be used for the development of fatherly making software. The proposed model is proposed is expected to produce
quality software is without defect. Of several previous studies there has been no accurate model for prediction of Software Defect due to the number of variables are many and varied which resulted in less accurate predictions. Model is good enough to do software defect prediction is C4.5, MLP and Naïve Bayes. Several other researchers also tried to improve the accuracy of existing by variable selection are used. The research on genetic algorithm will be applied to the selection of variables in methods of C4.5, MLP and Naïve Bayes using
data from the NASA dataset. After that will be tested with ROC curves and Confusion Matrix to find which model produces the highest level of accuracy in the prediction of software defects. Accuracy results obtained prove that the Naïve Bayes has a higher degree of accuracy than C4.5 and MLP. Naïve Bayes with Genetic Algorithm produces an accuracy percentage of 88.25% and the value of AUC (Area Under Curve) of 0.772. Thus Naïve Bayes algorithm optimized with a genetic algorithm to predict the Software Defect better