Xiao, Chunjing; Du, Xianghe; Song, Xueru; Xue, Yuxia; Wu, Minghao; Chetty, Kevin; (2026) Iterative feedback-based time-series anomaly detection with adaptive diffusion models. Neural Networks , 196 , Article 108370. 10.1016/j.neunet.2025.108370.

Text Chetty_Iterative_AnomalyTimes_accepted.pdf - Accepted Version Access restricted to UCL open access staff until 6 December 2026. Download (388kB)

Abstract

Anomaly detection in time series data is crucial across various applications. To enhance detection performance, imputation techniques, which follow a scheme of observed point selection → masked value estimation → anomaly determination, have been employed to capture complex correlations in time series data using advanced diffusion models. However, these imputation-based methods heavily rely on the user expertise and might suffer from performance degradation due to data distortion during the imputation process. To address these issues, we propose an Iterative Feedback-based Anomaly Detection framework with adaptive diffusion, IFAD. In this framework, we introduce an iterative feedback-based point selection scheme that identifies suitable normal points without depending on user experience. Further, we develop an adaptive conditional diffusion model with a dynamic weight-based data smoothing strategy, which can adjust the importance of observed points during the imputation process to generate smoothed data and enhance detection performance. Experimental results demonstrate that IFAD achieves significant improvements over state-of-the-art methods.

Download activity - last month

Export as XMLJSONCSV

Download activity - last 12 months

Export as XMLCSVJSON

Downloads by country - last 12 months

Export as JSONCSVXML

Archive Staff Only

View Item