Research question
Can anomaly detection be trusted when the monitoring data supplied to the model is itself incomplete, duplicated, or inconsistent?
Core idea
Model evaluation and data-quality evaluation should not be separate operational concerns. Pre-model validation creates an auditable boundary, while anomaly detection and explanation help distinguish unusual behavior from unreliable evidence.
Publication
A Cloud-Native Approach to Statistical Anomaly Detection and Automated Data Quality Validation. Chalapathi Koneni, Sanjay Lokula, and Lalan Panjiyar. International Journal of Data Science and IoT Management System, 4(4), 533–543, 2025.
Why it matters
A strong classifier score does not establish operational reliability. Teams also need to know whether the features were complete, whether validation ran, and whether the explanation is stable enough to support action.
Where the work goes next
Useful extensions include concept-drift handling, adaptive thresholds, cross-environment evaluation, and cost-aware decisions about when to investigate, throttle, or isolate a workload.