Reducing Features to Improve Code Change-Based Bug Prediction

Abstract

Abstract-Machine learning classifiers have recently emerged as a way to predict the introduction of bugs in changes made to source code files. The classifier is first trained on software history, and then used to predict if an impending change causes a bug. Drawbacks of existing classifier-based bug prediction techniques are insufficient performance for practical use and slow prediction times due to a large number of machine learned features. This paper investigates multiple feature selection techniques that are generally applicable to classification-based bug prediction methods. The techniques discard less important features until optimal classification performance is reached. The total number of features used for training is substantially reduced, often to less than 10 percent of the original. The performance of Naive Bayes and Support Vector Machine (SVM) classifiers when using this technique is characterized on 11 software projects. Naive Bayes using feature selection provides significant improvement in buggy F-measure (21 percent improvement) over prior change classification bug prediction results (by the second and fourth authors [28]). The SVM's improvement in buggy F-measure is 9 percent. Interestingly, an analysis of performance for varying numbers of features shows that strong performance is achieved at even 1 percent of the original number of features.

An Empirical Study on the Equivalence and Stability of Feature Selection for Noisy Software Defect Data

“…Song et al [3] pointed out that feature selection is an indispensable part of a general defect prediction framework. Shivaji et al [9] investigate the impact of six methods on the classification-based bug prediction. They found that selecting about 10% features could achieve a satisfactory performance.…”

An Approach for Predicting Bug Triage using Data Reduction Methods

“…The four well-performed algorithms are chosen in text data [13,19] and software data, namely Information Gain (IG), χ2 statistic (CH) [14], Symmetrical Uncertainty attribute evaluation (SU), and Relief-F Attribute selection (RF). The chi-squared distribution also known as chi-square or χ² distribution with k degrees of freedom is the distribution of a sum of the squares of k independent criterion normal random variables.…”

“…Shivaji and colleagues [14] proposed the feature selection techniques to predict the software bugs. Fu.Y, Zhu.X, and Li.B [7] investigated to obtain the accurate prediction model with minimum cost by labelling most informative instances.…”

Bug Triage Using Dimensionality Reduction Technique And PSO Algorithm

“…To enhance the data quality, Khoshgoftaar et al [11] and Gao et al [7] inspect the methods on feature selection to manage imbalanced defect records. Shivaji et al [23] suggests a structure to inspect multiple feature selection algorithms and eliminate noise features in classification-based defect prediction. Besides feature selection in defect prediction, Kim et al [13] present how to measure the noise confrontation in defect prediction and how to perceive noise data.…”