Enhanced OBKNN : An efficient framework for spectral data streams anomaly detection.

Computational efficiency is a key consideration in machine learning, particularly when real-time and accurate decisions are required. This challenge is even more pronounced in online anomaly detection, where continuously arriving data must be processed promptly, making detection latency and computational cost critical factors. We investigate in this work, the efficiency improvement of OBKNN, a spectral anomaly detection method, through the integration of the HNSW approximate nearest-neighbor search algorithm, as a promising enhancement aimed at significantly optimizing computational cost and thereby potentially offering sustainability benefits, achieving speedups exceeding a factor of nine (09) and a gain of 90\% in scoring time, while preserving the detection performance of the baseline method.