"Federated Medical Scoring Systems" by Jacob F. Bryant

Date of Award

3-2024

Document Type

Thesis

Degree Name

Master of Science in Operations Research

Department

Department of Operational Sciences

First Advisor

Chancellor Johnstone, PhD

Abstract

Federated Learning (FL) is a recent framework of machine learning implementation that trains models on a distributed network of clients as opposed to housing and analyzing this data centrally. This has data communication and practical data privacy advantages, the latter of which is particularly attractive to the medical community where patient privacy is closely safeguarded. We apply FL to a family of sparse linear integer models called Medical Scoring Systems (MSSs). We create a novel methodology for creating these MSSs in a simulated federated environment that involves an lo constrained Logistic Regression (LR), loss-surface examination, and rounding procedure. We tested this methodology on two datasets from the University of California Irvine’s Machine Learning Repository, the Heart Disease and Mushroom datasets. We evaluated our federated MSSs to six other model architectures of varying interpretability, sparsity, and centrality. Ultimately, we succeeded in creating a methodology for a federated MSS that addresses medical privacy concerns and enhances interpretability whose performance was similar to the highest performing centralized model.

AFIT Designator

AFIT-ENS-MS-24-M-069

Comments

A 12-month embargo was observed for posting this work on AFIT Scholar.

Distribution Statement A, Approved for Public Release. PA case number on file.

Share

COinS