Ml-based Anomaly Detection In Healthcare Systems Using Esp32 Microcontroller

Abstract

Continuous Monitoring Of Patient Vital Signs Through Wireless Body Area Networks (WBANs) Is Increasingly Central To Modern Healthcare, Yet Validating Anomaly-detection Pipelines On Real Hospital Networks Is Costly, Ethically Constrained, And Difficult To Reproduce. This Project Presents A Simulation-first Framework For ML-based Anomaly Detection In Healthcare Systems, Built Entirely In OMNeT++ With The INET Framework. The System Models Sensor Nodes (ECG, SpO2, Blood Pressure, Temperature, And Glucose), A Gateway, And A Central Server As A Discrete-event WBAN, In Which A VitalSignGenerator Module Produces Realistic Vital-sign Traffic And An AnomalyInjector Module Introduces Controlled Anomalies — Physiologically Abnormal Readings, Sensor Faults, Spoofed Packets, And DoS-style Flooding — At A Known, Configurable Probability. Because The Anomaly-injection Rate Is A Simulation Parameter, Every Generated Reading Carries An Automatic Ground-truth Label, Removing The Need For Manual Annotation. Simulation Results Are Exported Via Scavetool Into A CSV Pipeline And Used To Train And Compare Three Unsupervised Machine-learning Models — Isolation Forest, One-Class SVM, And An LSTM Autoencoder — Evaluated Against The Known Ground Truth Using Precision, Recall, F1-score, And ROC-AUC. The Framework Further Defines A Physical Sensor-hardware Layer (AD8232, MAX30100/MAX30102, MLX90614, And MPU6050 Interfaced To An ESP32 Microcontroller) As The Real-world Counterpart Of The Simulated WirelessHost Nodes, And Outlines Two Patterns — A Socket Bridge And An Embedded ONNX Runtime — For Optionally Closing The Loop With Real-time, In-simulation Detection. The Proposed Approach Offers A Reproducible, Extensible, And Ground-truth-evaluated Path From A Blank Simulation Workspace To A Deployable Anomaly Detector, Addressing Key Gaps In Existing WBAN And IoT Anomaly-detection Literature.

Keywords