Deep Learning–Based Classification of PPG-Derived Arterial Stiffness for Embedded Point-of-Care Systems

This dataset contains all numerical data used to generate the figures and tables reported in the manuscript entitled “PPG-Derived Arterial Stiffness Index Classification Using Deep Learning for Embedded Point-of-Care Systems.” The data were obtained from experimentally acquired photoplethysmography (PPG) signals and subsequent deep learning analysis. The dataset is provided in both Excel (.xlsx) and CSV (.csv) formats to ensure accessibility and reproducibility. Contents of the Dataset Figure 2 – Class Distribution File/Sheet: Fig2_Class_Distribution Description:Contains the number of samples in each arterial stiffness category (Normal, Mild, Severe) before and after training-set augmentation. Purpose:Supports analysis of class imbalance and the effect of controlled data augmentation. Figure 4 – Effect of Training-Set Augmentation File/Sheet: Fig4_Augmentation_Effect Description:Reports F1-scores of CNN, LSTM, BiLSTM, and GRU models with and without training-set augmentation. Purpose:Demonstrates the impact of augmentation on classification performance. Figure 5 / Table V – Consolidated Model Performance File/Sheet: Fig5_Consolidated_Performance Description:Includes accuracy, precision, recall, F1-score, average inference latency, and model size for all evaluated deep learning architectures. Purpose:Provides a comprehensive quantitative comparison of model performance and embedded feasibility. Figure 6 / Table VI – Class-Wise Performance File/Sheet: Fig6_Classwise_Performance Description:Contains precision, recall, and F1-score for each arterial stiffness class (Normal, Mild, Severe) for the selected model. Purpose:Supports class-wise performance analysis and discussion of classification behavior. Figure 7 – Robustness to Noise File/Sheet: Fig7_Noise_Robustness Description:Reports F1-scores under different levels of additive noise (low, medium, high). Purpose:Evaluates robustness of the proposed framework under realistic signal degradation. Data Format and Accessibility Formats provided: Excel (.xlsx) and CSV (.csv) Each figure corresponds to a clearly labeled sheet/file. No proprietary software is required to access the data. Reproducibility Statement All data in this repository were directly used to generate the figures and tables in the associated manuscript. The dataset enables independent verification of reported results and supports reproducibility of the experimental findings.