Statistical Parameter Optimization for On-Device PPG Artifact Detection under TinyML Constraints: A Multi-Class, Multi-Severity Analysis.

This repository contains the supporting materials for the research "Statistical Parameter Optimization for On-Device PPG Artifact Detection under TinyML Constraints: A Multi-Class, Multi-Severity Analysis." The study investigates the statistical influence of signal-processing parameters, feature configurations, and machine learning hyperparameters on photoplethysmography (PPG) artifact detection under resource-constrained TinyML deployment. The repository includes the feature-level dataset, data preprocessing scripts, parameter optimization framework, model training and evaluation code, statistical analysis scripts (including ANOVA and Pareto optimization), TensorFlow Lite (INT8) models, and embedded deployment examples for microcontroller-based wearable systems. These resources are intended to facilitate the reproducibility of the experimental results and support further research in wearable healthcare, biomedical signal processing, and embedded artificial intelligence. To protect participant privacy and comply with the approved ethics protocol, raw PPG recordings and personally identifiable information are not publicly distributed. The publicly available dataset contains de-identified feature representations used in the experiments. Additional data may be provided upon reasonable request and subject to institutional ethics approval.