PPGAIHI: Efficient Authentication in Healthcare IoT via CNN- and Distance-Based Thresholding

With the proliferation of devices connected through the Internet of Things (IoT), the focus on biometric authentication (BA) has intensified, driven by the increasing demand for convenience and security. Photoplethysmography (PPG) signals are recognized for their high usability, emerging as a viable method for user identification and authentication in remote medical environments. This study introduces PPGAIHI, a PPG-based authentication in healthcare IoT designed for practical applications and utilizing 1-D convolutional neural networks for feature extraction. Thorough performance evaluations of both model- and distance-based recognition methods were conducted. Comparative analyses of various PPG-based techniques were performed across ideal and real-world conditions using three distinct datasets to assess the tradeoff between performance and learning/recognition time, leading to system configuration recommendations. In ideal conditions, model-based methods, such as the support vector machine (SVM) and shallow fully connected networks achieved perfect scores-similarly, all distance-based methods achieved a score of one, contingent upon the selected threshold. In practical scenarios, the SVM achieved optimal accuracy, reaching one with more than six segments during the recognition phase. Conversely, the distance-based Minkowski model achieved a superior accuracy of 0.9779 using 100 segments (160 s) for registration and 15 segments (24 s) for recognition. Furthermore, this investigation provides practical guidelines for selecting optimal thresholds in distance-based recognition tasks, significantly improving the feasibility of PPG-based BA systems in real-world applications. This research demonstrates significant advancements in developing adaptable and practical PPG-based BA systems, achieving high accuracy under both ideal and practical conditions while effectively balancing system performance and recognition time.