Dynamic Anomaly-Missingness Fusion: Joint Detection of Data Deviations and Absences in Real-Time Systems

Current methods—like HM-LSTM for water quality (Tharayil et al., 2024) or moving-window z-scores for traffic data (Wang et al., 2023)—treat anomalies and missingness separately. However, in many systems (e.g., IoT, finance, healthcare sensors), these phenomena are intertwined: missing data often co-occurs with or signals underlying anomalies (sensor failure, fraud, attack). This research would propose a novel hybrid framework (potentially inspired by GAN-VAE models from Sekhar et al., 2025) that learns the joint distribution of normal, anomalous, and missing patterns, exploiting their correlations for more robust detection in real time. For example, abrupt increases in missing data might be causally linked to anomalous system behaviors. By fusing these signals, the algorithm could adaptively trigger alerts or impute values, outperforming models that treat missingness and anomalies independently. This approach has transformative potential for critical infrastructure, finance, and medical monitoring.

References:

1. A Methodology to Detect Traffic Data Anomalies in Automated Traffic Signal Performance Measures. Bangyu Wang, Grant G. Schultz, Gregory S. Macfarlane, Dennis L. Eggett, Matthew C. Davis (2023). Future Transportation.
2. Detecting Anomalies in Water Quality Monitoring Using Deep Learning. Sarafudheen M. Tharayil, Nada K. Alomari, Dana K. Bubshait (2024). Day 2 Wed, March 06, 2024.
3. EV Charging Station Management using Model Predictive Control and Generative Adversarial Variational Autoencoders. P. C. Sekhar, Harinath. G, Mujahid K, Pavani D (2025). 2025 3rd International Conference on Data Science and Information System (ICDSIS).