Answers are grounded in this paper and cite their sources.
Wireless sensor systems increasingly serve as the sensing layer of Internet of Things (IoT), industrial IoT and critical infrastructure applications, where corrupted readings can poison downstream classifiers, anomaly detectors and decision‐support models. This paper presents a poisoning‐aware data cleansing and restoration framework for secure IoT‐enabled wireless sensor systems. The framework treats suspicious readings as potential adversarial data‐poisoning attempts rather than solely as benign faults. Mobile Data Collectors (MDCs), static gateways or cluster heads act as edge sanitisation points that detect relationally inconsistent readings, neutralise suspicious values, restore them using co‐appearance‐, correlation‐ and similarity‐aware imputation, and produce anomaly reports for security‐aware decision support. The paper strengthens co‐appearance‐based cleansing through a richer threat model, robust statistical interpretation, optimisation‐based trimmed‐loss logic and adaptive thresholding. Using the Intel Lab WSN dataset, the framework is evaluated across 96 poisoning scenarios that cover different poisoning levels, perturbation ranges, distribution models and poisoning patterns. Results show a conservative, precision‐oriented detection profile, with an average Error Recall of 0.130 and Error Precision of 0.514, while restoration achieves an index of agreement of 0.888 and an RMSE of 0.121. The framework is best interpreted as a high‐confidence edge sanitisation layer within a defence‐in‐depth strategy.
No takes yet. Share an insight, caveat, or question.
Quazi Mamun (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: