Secure Medical Authentication for IoT-Enabled Healthcare Using Hybrid Deep Learning and Blockchain-Based Audit Logging

Abstract

As the number of Internet of Things (IoT) devices being used in healthcare grows, they provide a means for continuous monitoring of physiological measurements and network-level activities, but also introduce the requirement for reliable authentication, anomaly detection, and trustworthy event logging. This research proposes an authentication mechanism for health care systems based on the use of a secure method with Feedforward Neural Network (FNN) and Convolutional Neural Network (CNN) branches, and with an audit logging based on blockchain technology. The FNN branch aims at learning direct relationships between the heterogeneous physiological and network-related features, while the CNN branch is explored for learning on higher-order feature interactions following a structured transformation of the tabular representation. For class imbalance, Synthetic Minority Oversampling Technique (SMOTE) is not only added to the training set, but also to the training set only, thus avoiding information leakage to the validation and test sets. The prediction results are then added to a blockchain layer where the authentication-event metadata are turned into a stream of hash records and reported via smart-contract governed transactions for traceability and tamper evident auditing. The experimental protocol compares the proposed framework with individual FNN, CNN, DNN and LSTM baselines based on accuracy, precision and recall and F1 score, and area under the Receiver Operating Characteristic (ROC) curve. Other evaluations consider the cross-validation variability, confidence intervals, ablation analysis and the other performance measures related to the blockchain. The framework aims to offer a standardized solution for integrating the authentication with verifiable security logging in healthcare settings with IoT devices. The revised experimental design also reveals the restrictions of the small size of datasets, the representation of CNN in tabular form, computational complexity, and the actual deployment of blockchain systems, to serve as a more reproducible ground for the evaluation of the applicability of the proposed approach.