RESEARCH ON ANOMALY DETECTION METHODS IN API LOGS TO ENSURE SECURITY AND RELIABILITY OF SOFTWARE SYSTEMS

Abstract

With the increasing number of logs from various APIs, manual inspection and analysis is becoming an increasingly difficult task. Machine learning methods allow automating the process of analyzing large amounts of data and detecting unusual patterns that may indicate some anomalies or threats. Examining logs from API systems allows you to determine whether requests to the system are safe or suspicious. Anomalies in requests may indicate attempts at unauthorized access or malicious actions to a computer system. Today, companies often engage third-party pentesting specialists to periodically test software for vulnerabilities. However, to increase autonomy and security, it is advisable to implement a system that can independently identify abnormal traffic that may indicate potential threats. Such an approach will allow preventing attacks in advance and minimizing risks faster than waiting for results from external experts. The object of research is the processes of detecting anomalies in API logs to improve the security and reliability of software systems. The aim of the study is to investigate and evaluate the effectiveness of methods for detecting anomalies in API logs in order to improve the security and reliability of software systems. The study is aimed at comparing different unsupervised learning models and determining the most effective one for detecting potentially malicious activity in API logs. The paper investigates methods for detecting anomalies in API logs, which is critical to ensuring the security and reliability of software systems.Modern approaches to log analysis, including machine learning, statistical analysis, and anomaly detection methods, are considered. It is established that effective anomaly detection allows timely identification of potential threats, such as cyber attacks, system errors or unauthorized access, which significantly increases the level of software security.