An Interactive AutoML and LLM-Based Platform for Medical Data Analysis

Abstract

The widespread adoption of machine learning (ML) in healthcare is often limited by the significant technical expertise required to build, optimize and interpret predictive models. This paper presents an interactive platform designed to democratize access to medical data analysis by integrating an Automated Machine Learning (AutoML) engine with a Large Language Model (LLM) conversational assistant. The proposed system enables non-technical users, such as clinicians and biomedical researchers, to upload datasets and generate robust predictive models using AutoGluon without the need for programming. To address the interpretability challenges of complex models, the platformcouples SHAP-based visualizations with an LLM-driven chat interface that explains preprocessing steps, model metrics and feature importance in natural language. We evaluated the platform on eight public biomedical datasets covering both classification and regression tasks. Experimental results demonstrate that the system produces competitive predictive performance relative to established benchmarks while ensuring high usability. By lowering technical barriers and enhancing transparency, this tool empowers domain experts to independently leverage ML for clinical decision support and research.