ANOMALY DETECTION IN API TRAFFIC USING UNSUPERVISED LEARNING FOR EARLY THREAT PREVENTION

The growing complexity and volume of API-based communication in modern web services have made API gateways increasingly vulnerable to attacks such as abuse, fraud, and volumetric threats. Traditional rule-based or signature-based detection methods struggle to identify novel or evolving attack patterns in real time. This paper proposes an unsupervised learning-based framework for early anomaly detection in API traffic to address these limitations. Leveraging clustering algorithms and autoencoders, the system learns the normal patterns of API usage without labeled data and flags deviations as potential threats. The approach is designed to be protocol-agnostic and scalable across diverse microservice architectures. Empirical evaluation using real-world API traffic datasets shows that our method achieves high detection accuracy and low false positive rates while significantly reducing manual configuration effort. The findings suggest that unsupervised learning is a promising direction for proactive, adaptive API threat detection.