Machine Learning Algorithms Comparison for Software Testing Errors Classification Automation

Abstract

Automatic classification of software errors is an important tool for ensuring the quality of software being tested. By automating this process, testers can save time and effort, improve accuracy, gain a better understanding of errors, and improve communication between stakeholders. Software error clustering is an important concept in software testing that involves identifying and analyzing patterns in software errors or defects. The goal of error clustering is to understand the relationships between defects and identify the root causes of software errors, so that they can be prevented or corrected. The article provides a comprehensive survey of various techniques for software error clustering, including clustering algorithms. The author note that many of these techniques require significant expertise in software engineering and data analysis, and that there is a need for more user-friendly tools for software error clustering. This article presents an empirical research of software defect clustering, in which the author analyzes a large dataset of defects from a software development projects to identify several patterns in the defects, including clusters of related defects and common root causes.

The proposed software method uses stack traces to cluster data about software testing errors. The method uses a kNN algorithm to analyze test results, allowing the user to assign the text of the software test result to specified categories. The kNN algorithm has a high accuracy rate of 0.98, which is better than other clustering methods like Support Vector and Naive Bayes. The proposed method has several advantages, including a single repository for test results, automatic analysis of software testing results, and the ability to create custom error types and subtypes for error clustering. The developed new software method allows multiple launches to be combined. If there are a large number of test suites, they are divided into smaller groups as they cannot be included in a single run. The test suites can then be combined into a single run to present the data on dashboards and generate reports. The method is integrated into software using a Docker container, which reduces the time and human resources required for error analysis. Overall, the method provides real-time monitoring of project status for managers and customers.