An enhanced time series anomaly detection model based on frequency domain analysis

In recent years, detecting anomalies in time series data has an important function in industries including industrial monitoring, banking, and healthcare, and has attracted a lot of concern. Time series data typically exhibit varying intra and inter series correlations, leading to complex and intertwined dependencies, which traditional methods often struggle to capture, resulting in poor performance when dealing with complex timevarying patterns. This work presents an improved model for anomaly detection in time series data that leverages frequency domain analysis and element-by-element multiplication operations, and introducing Convolutional Block Attention Module(CBAM) to further optimize feature representation, thereby effectively enhancing the model’s capability to capture intricate temporal variations. Comprehensive experiments on two public datasets, NAB and UCR, show that the model we proposed significantly surpasses existing models. The improved model exhibits notable improvements in F1 score, recall, and precision, underscoring its capability to improve time series anomaly detection models’ performance.