10/01/2025
By Nazim Ahmed Belabbaci

Candidate Name: Nazim Belabbaci
Degree: Doctoral
Defense Date: Thursday, Oct. 9, 2025
Time: Noon - 1 p.m. EST
Location: Hybrid DAN 309 or via Zoom.

Committee Members:

  • Chair: Mohammad Arif Ul Alam, PhD, Assistant Professor, Richard A Miner School of Computer and Information Sciences, UMass Lowell
  • Benyuan Liu, PhD, Professor, Richard A Miner School of Computer and Information Sciences, UMass Lowell
  • Yan Luo, Ph.D. , Professor, Department of Electrical and Computer Engineering, UMass Lowell
  • Hassan Ghasemzadeh, PhD,  Program Director & Associate Professor, College of Health Solutions, Arizona State University (ASU)

Title: Integrated Hardware-Enabled Predictive Framework: Advancing Health Monitoring and Disease Diagnosis with Conditional Diffusion Models and Deep Learning

Abstract: "Accurate and continuous monitoring of health vitals, along with disease diagnosis from sequential data such as time-series and spectroscopy, is critical for health safety and optimizing human performance across diverse populations. Yet current methods often suffer from limited precision, insufficient data availability, and poor generalizability to real-world conditions. Many existing approaches depend on complex hardware or employ basic algorithms that fail to fully leverage the richness of temporal and spectral signals. Some methods are also further handicapped by a lack of robust, high-quality training data, ultimately limiting their effectiveness and reliability in practical, everyday healthcare applications.

This PhD research addresses these gaps across three application domains. First, in crowd health monitoring, we designed and validated hybrid predictive pipelines (Positionally Encoded Transformers, LSTM-Tabnet) on real-world physiological time-series data. This established a foundation for robust temporal modeling. Second, to address data scarcity in spectroscopy, we applied a class-conditioned diffusion model to FTIR spectral data for cancer diagnosis. We generated realistic synthetic spectra and combining them with the hybrid pipelines from the first phase to improve classification accuracy by up to 6%. Finally, in wearable hydration monitoring, we developed the HydroTrack prototype equipped with light spectroscopy sensors, completed an IRB-approved data collection. We aim to extend the end-to-end framework from sensing to synthetic augmentation and predictive modeling.

Collectively, these contributions demonstrate how we can overcome major barriers in health data acquisition and model generalizability. The proposed framework have strong potential to establish scalable, reliable, and data-efficient standards for health monitoring across diverse biomedical domains".