Edge-Assisted IoT Framework Detects Water Quality Anomalies in Bangladesh

Summary

A preprint from a Bangladeshi engineering university proposes an ESP32 + Raspberry Pi 4 edge stack that runs the Isolation Forest algorithm to detect anomalies in four water quality parameters (temperature, dissolved oxygen, pH, turbidity) in aquaculture. The study reported 97.25% accuracy on the IoTMLCQ dataset.

Why it matters

Real-time, low-cost anomaly detection at the edge could significantly reduce fish mortality in small-scale Bangladeshi aquaculture, where cloud-based data exchange is often impractical. However, this is a preprint benchmarking a standard algorithm on a public dataset. Therefore, its practical transferability to noisy, multi-farm field conditions remains unproven.

Highlights

• Both precision and recall are around 66%. This means that approximately one-third of true anomalies are either missed or flagged as false positives. This is a significant deficiency for an 'early warning' system.
• All evaluation was conducted on the public IoTMLCQ dataset. No live deployment, sensor drift analysis, or multi-farm validation was reported.
• The 'edge-enabled' architecture (ESP32 running Isolation Forest → Raspberry Pi 4) is a simple, common setup with limited algorithmic novelty.

Critical take

The 97.25% accuracy figure in the title is misleading, especially given the 66% F1 score. In an anomaly detection task with class imbalance, accuracy is nearly meaningless. It is difficult to characterize a system that misses approximately 33% of true anomalies as a reliable early warning tool. This peer-reviewed preprint, which benchmarks a standard algorithm on a single public dataset, reads more like a homework-level system integration than a research advancement.