Work

Developing a symptom-based disease prediction system using Machine Learning and Explainable AI (XAI), with an AI agent for interactive diagnosis assistance

Year: 2025
Type: dissertation
Abstract: The growing sophistication of medical diagnostics demands novel tools that will be used in clinical decision making. Whereas machine learning is an extremely effective tool at predicting trends, adoption has been negatively affected by limited transparency—the black box issue. This thesis meets this challenge with a comprehensive, but symptom-based mechanistic disease prediction system that combines high-accuracy modeling with Explainable AI (XAI) and an interactive AI agent. This project was used to train and test the machine learning models on a comprehensive dataset and a Logistic Regression classifier with accuracy of 86.7% was reported as the best predictor. Clinical trust and utility was assured by integrating a Local Interpretable Model-agnostic Explanations (LIME) framework, which explains the reasoning behind each diagnosis with transparent, medically intuitive explanations that key dependent symptoms. The main contribution of this work is the end-to-end system, that is orchestrated by a prototype AI agent which enables an interactive diagnostic dialog between the patient and the system. The involved agent efficiently collects user data, addresses the predictive model, and delivers diagnosis, along with its explanation, in a synthesized and human readable format. This produced proof-of-concept expresses a convincing, however, generalizable paradigm of developing reliable, explainable, and user-friendly AI-based clinical decision support systems.
Source: NORMA
Language: English
Open Access status: green
OpenAlex ID: W7204672647
Cites: 11
Cited by:
Related to: 10