Artificial Intelligence for Diabetes Disease

Abstract

This project presents an intelligent diabetes prediction and insulin dose estimation system, integrating artificial intelligence techniques to improve diabetes management. Based on the Indian PIMA diabetes database and medical recommendations, the system uses a neural network for binary classification (diabetic or non-diabetic) and a Random Forest regressor to calculate daily insulin doses (bolus and basal). The user-friendly interface, developed with Streamlit, allows both patients and practitioners to interact with the system. Clinical validation is carried out by a ratio test to assess the adequacy of insulin doses based on glycemic stability. The results confirm the reliability of the system and its potential to support early diagnosis and personalized treatment strategies, thus contributing to better health outcomes for diabetic patients. It achieves an accuracy of 77.92% and an precision of 70.59%, confirming its effectiveness for early diagnosis and treatment personalization.