Hybrid Adversarial Networks For Anomaly Detection In IoT Healthcare Applications

Anomaly detection is where data instances are discovered that deviate majorly from most data instances. Detection of anomalies in healthcare is considered a continuous research topic when concerned with IoT devices, which is becoming an integral part of people’s lives. Most existing anomaly detection approaches are tailored to specific use cases, necessitating both a deep understanding of the technique and the context of its application. As a rapidly growing domain, the Internet of Things (IoT) presents numerous opportunities for implementing such data analysis methods. However, the inherent complexities of the IoT make this integration challenging. Anomalies in ECG data may lead to false alarms and disease prediction can become uncertain. In order to avoid this, anomalies in ECG data have to be detected and treated accordingly. This project proposes a hybrid model involving Variational Autoencoder (VAE) and Generative Adversarial Networks (GAN) for detecting anomalies in MIT-BIH Arrhythmia dataset that contains the ECG signals and their corresponding annotations.