A Diabetes Prediction Model Using Hybrid Machine Learning Algorithm

Abstract

Diabetes is a major worldwide health issue, stressing the importance of early diagnosis and care. Machine learning algorithms offer promising prospects for developing precise models to classify diabetes. By leveraging vast healthcare datasets, machine learning can uncover hidden insights and patterns, enabling healthcare professionals to make informed predictions about patient outcomes. Despite advancements, current methods for diabetes classification suffer from accuracy limitations. In this research, we provide a novel hybrid machine learning approach that combines support vector machine, decision tree, and random forest classifiers. To improve forecast accuracy, we extend our technique by using new parameters like as glucose levels, BMI, age, and insulin levels. We trained and validated the algorithm using the Pima Indian Diabetes dataset using holdout and k-fold cross-validation approaches. On the holdout set, the hybrid method produced an accuracy of 88.5%, while k-fold cross-validation yielded 90.1%. While decision tree and random forest classifiers yielded individual accuracies of 76.8% and 75.3%, respectively, we further evaluated the algorithm's performance using recall, precision, and F1 score metrics. These indicators are critical in the field of diabetes prediction as they provide insights into the algorithm's capacity to correctly detect true positive cases and reduce false positives. They highlight the algorithm's effectiveness in diabetes prediction, making it a significant tool for early detection and intervention.