Data-Driven Machine Learning Methods for Diabetes Risk Prediction

The study developed machine-learning models to predict type 2 diabetes risk using a dataset of 520 participants, focusing on symptoms like polyuria, polydipsia, and alopecia. Random Forest and K-Nearest Neighbor (K-NN) models were the most effective, achieving high accuracy rates of 98.59% and 99.22% with SMOTE and different validation techniques. The research highlighted the importance of data preprocessing and feature ranking, identifying polyuria and polydipsia as significant predictors. The study demonstrated that machine-learning models could effectively aid in early diabetes diagnosis and management, despite limitations in the dataset's health profile details.