Implementation and Analysis of Supervised Learning methods for Bugs Classification

Abstract

Classification of bugs or logs is a vital aspect in the development of high-quality products. Early detection of errors, frequent patterns showing anomalies and, timely rectification of errors reduces the risk of developing faulty software. The aim of proposed system is to integrate a machine learning based intelligent layer to the existing Automation framework. The focus of this work is to classify the logs by extracting the underlying error messages in logs and hence ease the work of developers to save time invested in analyzing the logs. The data set includes logs collected from the Automation framework that were reported during automation runs in the last few months. This paper aims at finding the optimal algorithm to classify the bugs. In this experiment, we examine Naïve Bayes, Multilayer perceptron, CNN, and a hybrid model with a combination of CNN + Naïve Bayes Algorithm along with feature extraction techniques. It is observed from the experiment, that the Multilayer perceptron network has the highest accuracy of 91 percent. This experiment shows that our proposed system can classify logs effectively into different class of failure-types.