Data Science
Adaptive Noise-Resilient Anomaly Detection for Drifting Sensor Streams
Code: ANX-RN-001
Develop anomaly detection that keeps working when sensor streams drift, recalibrate and change noise characteristics over time, instead of assuming a fixed, stationary baseline.
Why it matters
Real deployments drift: sensors degrade, environments change and calibration ages. A detector tuned on one week of data quietly fails the next, and a failing detector is worse than no detector because it is trusted. Adaptive detection keeps the system honest over its whole lifetime.
Research gap
Most anomaly detection methods assume a stationary distribution. How to separate genuine concept drift from gradually worsening sensor noise — and update the model accordingly without absorbing anomalies into the baseline — is still an open modelling problem.
Research questions
- 01 How can a detector tell benign concept drift apart from a genuine anomaly in progress?
- 02 What update rule keeps false positives stable as noise characteristics change over months?
- 03 How much historical data must be retained to stay robust without unbounded growth?
- 04 How do adaptive detectors compare with static baselines on long-horizon deployment data?
Suggested technologies
Required skills
Team and supervisor
2-3 researchers, supervised by Muhammad Qasim.
Suitable for: DS, AI, CS, Embedded.
Expected deliverables
- Adaptive detection framework for drifting sensor streams
- Open drift and noise benchmark with public evaluation harness
- Research paper on separating concept drift from sensor degradation
Work on this problem
Apply to the lab and this problem becomes your supervised research project.