A Novel Machine Learning-Based Framework for Predicting Diabetes in Healthcare

The rise in diabetes is considered the greatest global health issue; thus, the call for early and accurate predictions of the disease is a dire need in the health sector. The detection of the disease in its initial stages would allow patients to undergo timely interventions and appropriate treatments. A novel approach for the prediction of stages and severity of diabetes by utilizing various models within a machine learning framework, such as Support Vector Machine (SVM), Artificial Neural Network (ANN), Decision tree learning (DT), and Random Forest (RF) is proposed in this paper. The training data would involve the history of medical conditions, age, BMI, insulin, skin thickness, glucose level, the number of pregnancies, and existence or non-existence of diabetes. The models were trained on 70% of the data and tested on the remaining 30%. The results show that SVM outperforms all other models, achieving an accuracy of 96.5%, followed by ANN with 94.5%, DT with 90.2 %, and RF with 86.7%. These results indicate that SVM, followed by ANN, is the most effective model for predicting diabetes stages. The accuracy and reduction of data loss have been constant improvements with epochs 30 to 360; therefore, models can be used in real-time applications in the clinical area. Thus, this research concludes that machine learning models, such as SVM, may be helpful in predicting and diagnosing early cases of diabetes; thus, these models could be helpful in making good decisions on healthcare and improving the patient's health.