Continuous health and well being data management using wearable sensors: an integrated approach to data collection, processing, and analysis.

Continuous monitoring of cardiovascular dynamics offers a promising window into an individual’s health and wellbeing, yet wearable photoplethysmography (ppg) data present unique challenges in signal quality, reliability, and real time analytics. in this thesis, an end-to-end framework is introduced for the integrated management of continuous health and wellbeing data using wrist worn wearable sensors. twenty hour long (twenty-one-hour) recordings from ten healthy adults were collected with tri wavelength ppg and tri axial accelerometery raw samples were resampled to a uniform 25hz grid, cleaned via adaptive filters and signal quality indices, and segmented into non overlapping 30s windows. fifteen features time and frequency domain Heart Rate Variability (HRV) metrics and pulse morphology descriptors, were extracted and rigorously validated. reliability analyses (Cronbach’s α=0.46, test-retest r ≤ 0.63) revealed modest stability at 30s, which improved substantially when aggregating to 60s and 120s windows. Unsupervised Kmeans clustering uncovered four physiologically meaningful states parasympathetic dominant, sympathetic dominant, baseline survey, and vascular responsive, that aligned with reading, writing, discussion, call, and survey periods. principal component analysis further reduced dimensionality, with the first two components explaining ≥ 50 % of variance and loading heavily on SDNN, RMMSD, pulse amplitude, and pNN50. supervised classifiers (random forest, SVM) achieved only modest five-way activity recognition accuracy (≤ 40 %), underscoring the limits of short window decoding and motivating low dimensional or unsupervised representations. This work contributes, (Ⅰ) a validated open-source pipeline for PPG based feature extraction, (Ⅱ) an autonomic state model interpretable in terms of engagement and stress, and (Ⅲ) practical guidelines on window selection, feature prioritization, and on device processing for real world continuous health and well-being monitoring. future directions include longer aggregation windows, richer contextual fusion, and real-world trials to deliver personalized, context aware interventions.