MACHINE LEARNING ALGORITHMS FOR ANOMALY DETECTION IN PUBLIC DATA USING GITHUB AS AN EXAMPLE

Аннотация

INTRODUCTION project evaluations. The falsification of metrics, for GitHub [2], one of the largest platforms for hosting example, fake stars, creates a false impression of a and collaborative development of software, has project's popularity and significance, which can become an integral part of the modern mislead investors and users. As a result, identifying technological landscape. Its role in evaluating anomalies in such data becomes a critical task for technology projects and startups is growing as ensuring transparency and fairness in the more investors and developers turn to platform technology community [3]. activity metrics to make decisions on funding and collaboration [4]. However, with the increasing Machine learning (ML), as an advanced data popularity of GitHub, a significant issue has analysis technology, opens new possibilities for arisen—manipulation of metrics such as stars, detecting such anomalies. Machine learning commits, and forks, which calls into question the algorithms can process large amounts of data, reliability of these data as indicators of project identifying hidden patterns and deviations from success and quality [3]. the norm that are difficult to detect using traditional analysis methods. Applying these The relevance of studying methods for analyzing algorithms to GitHub data enables the automation public GitHub data is justified by the need to of detecting suspicious activities, such as sudden improve the accuracy and reliability of technology https://www.theamericanjournals.com/index.php/tajet 33 THE USA JOURNALS THE AMERICAN JOURNAL OF ENGINEERING AND TECHNOLOGY (ISSN – 2689-0984) VOLUME 06 ISSUE10 spikes in stars or unusual commit patterns, interest in a repository. Users can "star" contributing to a more accurate assessment of repositories to show their support or interest. projects. - Forks: These are copies of a repository created for The goal of this study is to develop a methodology further development or modification of the code. for detecting anomalies in GitHub data using Forks are an important indicator that a project is machine learning algorithms. The study will attracting developers and has potential for further explore the main approaches to analyzing such development. data, conduct a comparative analysis of various - Pull requests: These are proposals to make machine learning algorithms, and propose a changes to a repository, which can be accepted or conceptual model of a system for automated rejected by the project owners. Pull requests anomaly detection. This model can be applied to reflect community activity and the involvement of projects focused on evaluating technology external developers. startups, such as DualSpace AI, providing more Each of these elements can be analyzed to detect reliable and objective analysis results. anomalies that may indicate unnatural or Thus, the research is aimed at addressing the manipulative actions. important task of improving the transparency and Anomalies in GitHub data can take various forms, reliability of GitHub data, which ultimately each representing a potential threat to the contributes to better decision-making in software accuracy of project analysis and evaluation. It is development and investment. important to classify these anomalies to develop 1. Theoretical foundations and methodology effective methods for detecting and mitigating for detecting anomalies in GitHub data them. GitHub [2], as a platform for hosting and The main types of anomalies include: collaborative software development, provides a - Star anomalies: Sudden spikes in the number of wide range of public data that can serve as a basis stars may indicate manipulation, where stars are for analyzing and evaluating technology projects bought or generated by bots. These anomalies are [4]. These data include, but are not limited to, the easily noticeable when analyzing time series data, number of stars, forks, commits, pull requests, as revealing unnatural patterns. well as metadata about users, repositories, and project-related events. - Commit anomalies: Excessive activity in the form of numerous minor or irrelevant commits over a The data structure on GitHub is organized into short period may indicate attempts to create the repositories, each representing a container for illusion of active development. code and associated artifacts. The main elements of a repository include: - Fork anomalies: A sharp increase in the number of forks, especially if these forks do not lead to - Commits: These are fixed changes to the code that further activity, may signal manipulation aimed at are recorded in the change history. Commits increasing a project’s visibility. contain information about who made the changes, when they were made, and a description of the - Pull request anomalies: A large number of pull changes. requests from inactive or newly created accounts may indicate artificial activity designed to boost - Stars: This metric reflects the popularity a