LLM-Based Classification in Diabetes Disease Diagnosis for Health Consultancy

Abstract

Diabetes is a chronic disease that is rapidly spreading throughout the world and causes serious health problems. Early diagnosis is especially important to prevent the disease from progressing and causing different complications. In this study, ChatGPT and Gemini Large Language Models (LLM) were used in the diagnosis of diabetes. The models were tested with two different prompting techniques including Zero-Shot and Chain-of-Thought (CoT). The obtained results were evaluated in terms of Accuracy, Sensitivity, Specificity, Precision and F1-Score performance metrics, computation time and token cost. As a result of the analysis, it was seen that the ChatGPT model achieved the highest success rate with the Zero-Shot prompt technique and was also the lowest cost and fastest working model. These findings suggest that LLM-based health applications can be developed and contribute to the design of systems that can assist doctors.

Ähnliche Arbeiten