AI-Powered API Monitoring and Multi Source Anomaly Identification Model for Distributed Platforms

Abstract

APIs facilitate communication between distributed systems (cloud services, applications, microservices). However, APIs are subject to various forms of attack, compromise, or access via network communications due to their inherent nature as public interfaces or entry points into secure environments. As a result, traditional rule-based anomaly detection has difficulty identifying complex or unknown anomalies. In this paper, we propose an AI-Powered API Monitoring and Multi-Source Anomaly Identification Model leveraging Deep Learning with real-time capabilities to conduct anomaly identification. We utilized telemetry data collected from the Train Ticket Dataset provided in the AIOps Challenge 2020 event, consisting of API Logs, System Metrics, and Network Trace data. Collected telemetry data was analyzed using three classifiers (MSIF-LSTM, PLE-GRU, & Hybrid Ensemble). The Hybrid model was able to achieve an overall accuracy rate of 96.40%, demonstrating excellent scalability and reliability. Including real-time API monitoring and automated alerting capabilities, our proposed model improves API security and reliability while reducing the number of unknown or outlier threats to the distributed systems using APIs by providing a proactive approach to threat identification and prevention.