Penerapan Greedy Forward Selection dan Bagging pada Logistic Regression untuk Prediksi Cacat Perangkat Lunak

摘要

Software defects are errors or failures in software. Software defect detection manually can only produce 60% of the total existing defects. Defect prediction method using probability to find up to 71% better than the method used by the industry. One of the best methods for prediction of software defects is Logistic Regression. Logistic Regression is a linear classifier that has been shown to produce a powerful classification with statistical probabilities and handle multi-class classification problem. The main weakness of Logistic Regression algorithm is a class imbalance in high-dimensional datasets. Dataset software metric used is NASA dataset MDP. The dataset is generally unbalanced and experiencing problems with redundant data. This paper proposed a greedy forward selection method to solving the problem of redundant data and bagging technic to solving the class imbalance. The algorithm used is the Logistic Regression. Results of the experiments in this study scored the highest accuracy in the dataset PC2 at 0,990, up 0.19% compared with logistic regression method without GFS and bagging. While the highest AUC value of 0.995 at PC2, an increase of 7.94% compared to logistic regression method without GFS and bagging.