Design and development of an LLM-based framework for synthetic data generation

The increasing demand for high-quality datasets in fields such as healthcare, finance, and cybersecurity is hindered by challenges such as data scarcity, privacy concerns, and regulatory restrictions. This thesis introduces a novel framework for generating synthetic data using fine-tuned Large Language Models (LLMs) and Generative AI techniques. The framework generates realistic, domain-specific datasets that preserve complex patterns while ensuring privacy through differential privacy methods. It can create synthetic data from scratch and augment existing datasets, thereby offering a scalable solution across industries. Prototype implementation and extensive testing demonstrate the framework’s effectiveness in balancing data utility and privacy, making it a valuable tool for overcoming data access challenges while complying with privacy and regulatory standards. A comprehensive evaluation comparing proprietary and open-source LLMs demonstrates the framework’s superiority in terms of data fidelity, statistical similarity, and computational efficiency.

Name thesis:degree_name

Master of Applied Science (MASc)

Discipline thesis:degree_discipline

Electrical and Computer Engineering

Grantor

University of Ontario Institute of Technology

Year dc:date.issued

2025

Author dc:creator

• Goyal, Mandeep

Advisor dc:contributor.advisor

• Mahmoud, Qusay H.

Language dc:language.iso

en

Handle dc:identifier.uri

https://hdl.handle.net/10155/1930

OAI identifier oai:identifier

oai:ontariotechu.scholaris.ca:10155/1930