A Reliable Approach for Lightweight Anomaly Detection in Sensors Using Continuous Wavelet Transform and Vector Clustering

In the rapidly evolving field of sensor technology, efficient and accurate anomaly detection is critical across applications from environmental monitoring to cybersecurity. Traditional approaches often fail in real-time sensor data scenarios due to high computational requirements and lack of labeled datasets. This paper presents a lightweight, unsupervised anomaly detection framework that combines Continuous Wavelet Transform (CWT) with Support Vector Clustering (SVC), aiming to reduce computational complexity and dynamically adapt to the data flow. Extensive validation on the Intel Berkeley Research Laboratory (IBRL) dataset demonstrates that our method not only handles sensor aberrations effectively, but also achieves a significant detection accuracy of 93.2% for drift readings, confirming its robustness and efficiency.