Bug Classification: Feature Extraction and Comparison of Event Model using Naïve Bayes Approach
This work addresses bug classification for software engineers and testers, but it is incremental as it applies existing methods to a specific domain.
The paper tackles the problem of classifying network bugs by developing a structured mining algorithm that uses feature extraction, noise reduction, and a probabilistic Naïve Bayes approach with Bernoulli and Multinomial event models, resulting in performance comparisons based on accuracy and recall parameters for unseen bugs.
In software industries, individuals at different levels from customer to an engineer apply diverse mechanisms to detect to which class a particular bug should be allocated. Sometimes while a simple search in Internet might help, in many other cases a lot of effort is spent in analyzing the bug report to classify the bug. So there is a great need of a structured mining algorithm - where given a crash log, the existing bug database could be mined to find out the class to which the bug should be allocated. This would involve Mining patterns and applying different classification algorithms. This paper focuses on the feature extraction, noise reduction in data and classification of network bugs using probabilistic Naïve Bayes approach. Different event models like Bernoulli and Multinomial are applied on the extracted features. When new, unseen bugs are given as input to the algorithms, the performance comparison of different algorithms is done on the basis of accuracy and recall parameters.