Federated Learning-Enabled Adversarial Attack Detection Framework for Healthcare Laboratory Data

An adversarial attack deliberately manipulates data or model inputs to deceive a machine learning model, causing it to make incorrect predictions or classifications. In machine learning and artificial intelligence, adversarial attacks exploit vulnerabilities in models by subtly altering the input data in ways that are often imperceptible to humans but cause the model to misbehave. These attacks can target various machine learning models and healthcare applications. This paper presents a federated learning-enabled framework that collects and trains healthcare data from different laboratories and hospitals and aggregates it in a centralized hospital for disease prediction. We present the AADFLS (Adversarial Attack Detection Federated Learning Scheme), which consists of different steps to identify attacks and securely process data between connected hospitals and laboratories. We aim to collect and train data securely and avoid adversarial attacks in the distributed healthcare environment. Simulation results show that our testbed results are based on AADFLS having higher accuracy and scores as compared to existing adversarial attack detection schemes (AADS) and adversarial attack federated learning schemes (AAFL) when shared and aggregated data for training predictions on the aggregate node.