Apply multiple machine learning models to diabetes prediction

Abstract

Diabetes mellitus, a pervasive and chronic metabolic disorder, imposes a substantial burden on global health systems due to its requirement for lifelong management and the myriad of complications associated with inadequate control. The ability to accurately forecast the onset of this disease is paramount, as it enables preemptive interventions and tailored treatment strategies that can significantly mitigate its impact. This paper investigates the application of machine learning techniques and deep learning models in diabetes prediction. This paper makes use of the Pima Indian Diabetes Dataset (PIDD) from Kaggle, which has 768 data entries and eight characteristics like blood pressure, blood sugar, and body mass index (BMI). Various algorithms, including Support Vector Machine (SVM), Decision Trees(DT), Random Forest(RF) , and Fully Connected Neural Network (FCNN), are implemented and compared. Identifying the strengths and limitations of each model, the results emphasize the potential of advanced computational models in improving the accuracy and clinical usefulness of diabetes prediction. The best-performing model is the FCNN model, with a test accuracy of 78.67% and an AUC value of 83.36%.