iMIF2V2: Intelligent Multiple Imputation Federated Fuzzy Clustering with Visualization and Validation for Longitudinal Digital Trials
Dissertation Advisor: Dr. (Julia) Hua Fang - University of Massachusetts Dartmouth - CIS
Committee members:
Dr. Long Jiao; Dr. Honggang Wang; Dr. Gang Zhou, University of Massachusetts Dartmouth - CIS William & Mary
Abstract:
Artificial intelligence (AI)-driven behavior recognition and privacy-preserving machine learning frameworks offer strong potential for digital behavioral trials. Yet multi-site longitudinal studies face major challenges: incomplete and high-dimensional data, non-normal distributions, evolving trajectories, and strict privacy regulations (e.g., HIPAA), where even anonymized datasets may risk re-identification. Soft clustering is well-suited to these settings by allowing partial membership and modeling overlapping, dynamic trajectories. Coupled with multiple imputation, it provides a principled solution for incomplete longitudinal digital trial datasets. Existing methods, however, lack efficient soft encoder optimization and integrated clustering, validation, and visualization within a federated framework. To address this gap, Intelligent Multiple Imputation Federated Fuzzy Clustering with Visualization and Validation (iMIF2V2) is proposed, a decentralized AI algorithm that unifies soft encoder optimization, fuzzy cluster validation, and visualization for incomplete longitudinal data. Using weighted rank aggregation and 3D visualization, iMIF2V2 enables adaptive fuzzifier tuning, interpretable modeling of evolving behavior patterns, and federated learning across sites without sharing raw data. Validation includes ablation studies on simulated datasets (varying sites, clusters, patients, effect sizes, correlation structures) and empirical data: harmonized longitudinal dietary quality from four Massachusetts RCTs (n=957) and national studies (>3.3 million observations). Results show iMIF2V2 converges rapidly, identifies cluster numbers accurately, and achieves strong performance across indices, particularly with larger effect sizes and adequate site-level samples. It mitigates missingness and heterogeneity while requiring minimal communication rounds, ensuring practical and efficient federated implementation. A demonstration emulates federated settings with two GPU servers acting as clients, enabling users to select among multiple longitudinal trials for clustering. Outputs include local and global centroids, dietary trajectory plots, and optimized 3D/2D Sammon projections. By integrating intelligent fuzzy clustering, multiple imputation, visualization, and federated learning, iMIF2V2 may broaden opportunities for scalable, interpretable, and privacy-preserving analyses across multi-site longitudinal behavioral digital trials.
All EAS students are encouraged to attend and all interested parties invited.
For further information, please contact Dr. Hua Fang.
Zoom Online
Hua (Julia) Fang
hfang2@umassd.edu
https://umassd.zoom.us/j/95683754446?pwd=WW1paGw5Q29mdXBMb0E3N3dkUTZ2Zz09