Uncertainty-Aware Contamination Detection in IoT Water Networks via Interval Type-2 Fuzzy Rare Itemset Mining

Uncertainty in low-cost Internet of Things (IoT) sensors challenges real-time contamination identification in institutional water infrastructure. Typically, conventional threshold-based alert systems fail to detect anomalies across multiple correlated parameters that are not individually outside the 'safe' limits. We introduce T2MFRM (Type-2 Multiple Fuzzy Rare Itemset Mining), a framework that identifies unusual but significant contamination patterns using uncertainty mining. The system uses Interval Type-2 Fuzzy Sets (IT2FS) to capture sensor measurements under uncertainty by defining a Footprint of Uncertainty (FOU). To mine unusual, rare patterns, we use a hash-table structure with upper-bound pruning to handle the combinatorial explosion associated with mining rare patterns. This preserves computational efficiency for deployment on edge gateways. T2MFRM was benchmarked against FRI-Miner and RP-Growth on six publicly available datasets and ran significantly faster while consuming less memory. The framework's advantage increased as the maximum frequent-support threshold (minFS) grew from 20% to 36%. We also tested the framework on real-world datasets. One case study demonstrates that, contrary to systems which rely on single-parameter thresholds, T2MFRM detected a multivariate contamination signature in the form of a simultaneous excursion of pH and temperature, neither of which was individually outside of safe limits. The results showed that the proposed system is computationally efficient and increases sensitivity to pollution events in a water management system compared with the traditional Boolean logic approach.