Comparative Analysis of Machine Learning and Large Language Model Architectures for Intelligent Medical Diagnosis and Clinical Decision Support Systems
Year: 2026
Type: conference-paper
Abstract: Artificial Intelligence (AI) and Machine Learning (ML) have dramatically changed the methods involved with medical diagnostics and decision-making in the clinic, as well as the use of informatics within healthcare. In this paper, we present a rigorous comparison of six widely used and popular machine learning and artificial intelligence architectures in the creation of a clinical decision support/predictive analytic model: Random Forest, Support Vector Machine (SVM), Multilayer Perceptron (MLP), Gradient Boosting, BERT-based Transformer and GPT-4 fine-tuned model [1],[2],[3]. Evaluation of each model was conducted using a large-scale clinical dataset of 500,000 patients from multiple tertiary care hospitals throughout India, and was based on accuracy, precision, recall, F1-score, AUC-ROC, computational scalability and clinical interpretability. Our results indicate that the GPT-4 fine-tuned model achieved state-of-the-art classification accuracy of 97.2%, surpassing all other models tested. Additionally, we demonstrate a novel hybrid ensemble architecture (MediConnect-AI) that uses gradient-boosting classifiers with transformer-based contextual embeddings to achieve 98.1% classification accuracy with significantly reduced inference latency [4],[5],[6]. Using ablation studies, feature significance analysis, and extensive statistical significance testing, we demonstrate that our new framework is superior to all tested models. Collectively, these results will have a significant impact on the future deployment of AI systems in real-world clinical settings.
Institution: St. Andrews Institute of Technology and Management
Language: English
Subfield: Health Information Management
Field: Health Professions
Domain: Health Sciences
Open Access status: hybrid
OpenAlex ID: W7203878618