Comparative Assessment of Machine Learning Models for Predicting Glucose Intolerance Risk

• Published in: SN computer science 2024-09, Vol.5 (7), p.894, Article 894
• Main Authors: Kumar, B. P. Pradeep, Manoj, H. M.
• Format: Article
• Language: English
• Subjects: Accuracy; Advances in Computational Approaches for Image Processing; Algorithms; Clinical outcomes; Cloud Applications and Network Security; Comparative analysis; Computer Imaging; Computer Science; Computer Systems Organization and Communication Networks; Customization; Data mining; Data Structures and Information Theory; Datasets; Decision trees; Deep learning; Global health; Glucose; Health care; Information Systems and Communication Service; Lifestyles; Literature reviews; Machine learning; Medical research; Metabolic disorders; Pattern Recognition and Graphics; Performance evaluation; Precision medicine; Prediction models; Predictions; Predictive analytics; Public health; Review Article; Risk assessment; Risk factors; Software Engineering/Programming and Operating Systems; Vision; Voting; Wireless Networks
• ISSN: 2661-8907; 2662-995X; 2661-8907
• DOI: 10.1007/s42979-024-03259-5

This study focuses on predictive analytics and healthcare, specifically the prediction of Glucose Intolerance, a chronic metabolic disease with significant global health implications. The research aims to develop efficient tools for risk assessment and early identification, offering medical practitioners reliable instruments for identifying high-risk patients and implementing preventive measures in a timely manner. A significant challenge in managing Glucose Intolerance is the absence of effective and precise prediction models. Traditional risk assessment techniques often fail to capture the multifaceted nature of Glucose Intolerance development, leading to delayed treatments and suboptimal patient outcomes. To address this issue, we conducted a comprehensive study utilizing various machine learning algorithms, including Decision Trees, Random Forest, Gradient Boosting, CatBoost, K-Nearest Neighbors (KNN), Support Vector Classifier (SVC), Logistic Regression, and a Voting Classifier, to predict Glucose Intolerance. The primary objective was to identify the most effective combination of features and models for accurate predictions. Key factors considered included patient demographics, lifestyle characteristics, medical history, and genetic susceptibility, which were used to build robust and personalized prediction models. We conducted a comparative analysis of the machine learning models' performance based on cross-validated accuracy with test-train splits and folds: 0.2, 6; 0.05, 4; 0.1, 2; 0.2, 6; and 0.05, 2. The results showed that Random Forest achieved test accuracies of 74%, 76%, 82%, 75%, and 79%; KNN achieved 70%, 72%, 74%, 75%, and 72%; SVC achieved 74%, 72%, 77%, 77%, and 72%; Logistic Regression achieved 73%, 76%, 79%, 73%, and 76%; Gradient Boosting achieved 72%, 79%, 75%, 73%, and 79%; XGBoost achieved 72%, 79%, 77%, 74%, and 79%; CatBoost achieved 74%, 76%, 79%, 75%, and 76%; Decision Tree achieved 60%, 83%, 74%, 70%, and 79%; and Voting Classifier achieved 74%, 72%, 77%, 74%, and 72%. These results highlight the strengths and limitations of each model, providing insights into their suitability for Glucose Intolerance prediction. The findings have significant implications for the prevention and early detection of Glucose Intolerance. By accurately identifying high-risk patients, healthcare practitioners can implement targeted interventions and lifestyle modifications, thereby improving patient outcomes and reducing the overall burden of Gluco