@article{c426b8d0-ae89-41d3-99be-6c80175dc0af,
  abstract     = {{<p>Machine learning (ML) is increasingly being explored in medicine and healthcare to improve decision-making and predictive capabilities. However, further research is needed to effectively apply ML models to the reliable monitoring of infectious disease outbreaks using multivariate symptom surveillance data. A central component of such monitoring is anomaly detection. To address limitations in current anomaly detection systems and evaluation practices, we propose an epidemiology-guided machine learning framework for real-time anomaly detection in disease outbreak monitoring. We simulate three types of disease outbreaks (A, B, and C), designed to mimic real-world diseases, by modeling transmission dynamics using the SEIR (Susceptible–Exposed–Infectious–Recovered) epidemiological model. Our framework integrates reconstruction-based anomaly detection models and updates them over time, allowing detection to adapt as epidemiological patterns change. We evaluate the proposed framework on empirically observed multivariate time series augmented with outbreak ground truth. In addition, we introduce an event-centric evaluation metric based on density-based clustering to better capture temporally coherent anomaly patterns, and also analyze naturally occurring disease outbreaks in the data. Using this framework, we evaluate multiple anomaly detection models across different outbreak scenarios. Results indicate that the proposed Transformer-AE generally achieves competitive performance and frequently outperforms the other evaluated reconstruction-based models across the investigated scenarios. In general, performance declines with lower transmission rates, highlighting the increased difficulty of detecting mild or slowly emerging outbreaks in multivariate symptom surveillance data. Overall, the proposed framework provides a realistic and interpretable approach for evaluating reconstruction-based machine learning models for outbreak monitoring through adaptive learning and event-level assessment.</p>}},
  author       = {{Hashemi, Atiye Sadat and Dietler, Dominik and Ohlsson, Mattias and Björk, Jonas}},
  keywords     = {{Disease outbreaks; Early detection; Multivariate time series; Reconstruction-based models; Time-series anomaly detection}},
  language     = {{eng}},
  pages        = {{1--15}},
  publisher    = {{Elsevier}},
  series       = {{Machine Learning with Applications}},
  title        = {{An epidemiology-guided machine learning framework for real-time anomaly detection in disease outbreak monitoring using symptom surveillance data}},
  url          = {{http://dx.doi.org/10.1016/j.mlwa.2026.101015}},
  doi          = {{10.1016/j.mlwa.2026.101015}},
  volume       = {{26}},
  year         = {{2026}},
}