Multiclass Diabetes Detection Using Random Forest Classification

Abstract

Detecting diabetes at an early stage can help save lives and improve the patients quality of life significantly. Diabetes can be detected with the assistance of information regarding the patient's lifestyle and health. This work aims to predict diabetic patients using different machine-learning classification algorithms and a dataset about diabetic and healthy patients. The work employs a data balancing technique to handle the data imbalance issue, as well as using cross-validation. In addition, it compares these machine-learning algorithms according to several performance indicators like accuracy, precision, recall, and Fl-score. Accordingly, the Random Forest classifier proved to produce the best results with accuracy, precision, recall, and an Fl-score, all equal to 89%.