Evaluating the Performance of Ensemble Learning Methods in Diabetes Disease Classification

Evaluating the Performance of Ensemble Learning Methods in Diabetes Disease Classification | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Evaluating the Performance of Ensemble Learning Methods in Diabetes Disease Classification Sajjad aghasi javid, Aliasghar Khakpaki This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7775507/v1 This work is licensed under a CC BY 4.0 License Status: Under Revision Version 1 posted 10 You are reading this latest preprint version Abstract Diabetes mellitus is a prevalent metabolic disorder characterized by chronic hyperglycemia and associated with severe complications. Accurate early detection is essential for effective management and prevention of disease progression. This study systematically evaluates the performance of three ensemble learning approaches Bagging, Boosting, and Stacking on three benchmark diabetes datasets: Pima Indians Diabetes, Frankfurt Hospital Diabetes, and Sylhet Hospital Diabetes (NIDDK). Class imbalance, a common challenge in these datasets, was addressed using the Synthetic Minority Oversampling Technique (SMOTE) during preprocessing to enhance model stability and classification reliability. Experimental results indicate that Boosting-based methods consistently outperform Bagging and Stacking. On the Pima dataset, Gradient Boosting, Extreme Gradient Boosting, and CatBoost achieved a maximum accuracy of 81.82%. On the Frankfurt dataset, Light Gradient Boosting reached 99.25% accuracy, while on the NIDDK dataset, Light Gradient Boosting and CatBoost attained perfect accuracy (100%). These findings highlight the effectiveness of integrating SMOTE with Boosting-based ensemble models to mitigate class imbalance and improve diabetes classification. The results underscore the importance of both data preprocessing and algorithm selection in achieving high predictive performance, with significant implications for precision medicine and clinical decision support. Biological sciences/Computational biology and bioinformatics Health sciences/Diseases Health sciences/Endocrinology Health sciences/Health care Physical sciences/Mathematics and computing Health sciences/Medical research Bagging Boosting Stacking Diabetes Ensemble Learning Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Introduction Type 2 diabetes mellitus is one of the most prevalent metabolic disorders, characterized by chronic hyperglycemia, and is recognized as a major risk factor for coronary heart disease, stroke, and other cardiovascular complications 1 . The pathogenesis of diabetes primarily involves two key mechanisms: insufficient insulin secretion from pancreatic β-cells and reduced sensitivity of peripheral tissues to insulin. Insulin, as a critical hormone, plays a fundamental role in maintaining glucose homeostasis, and any dysfunction in its activity can lead to severe systemic complications affecting multiple organs, including the heart, kidneys, and nervous system 2 . Globally, diabetes prevalence is rising at an alarming rate, representing a significant public health challenge. According to the World Health Organization (WHO), the number of individuals diagnosed with diabetes increased from 108 million in 1980 to 422 million in 2014, with projections estimating over 1.3 billion by 2050 3 . This increase is disproportionately concentrated in low- and middle-income countries, highlighting significant disparities in healthcare access, preventive services, and disease management 4 . In recent years, medical research has increasingly focused on early detection, risk stratification, and disease classification. Machine Learning (ML) has emerged as an essential tool for analyzing high-dimensional and complex datasets, detecting subtle patterns, and developing accurate predictive and classification models 5 . A major challenge in medical datasets, particularly diabetes-related data, is class imbalance, where the number of healthy samples typically exceeds that of patient samples. This imbalance can significantly reduce model performance and increase prediction errors 6 . To address these challenges, data preprocessing techniques such as Synthetic Minority Oversampling Technique (SMOTE) and advanced ML algorithms have been widely adopted. Among these, Ensemble Learning methods including Bagging, Boosting, and Stacking have demonstrated superior performance in medical data classification. By combining multiple base models, ensemble approaches reduce generalization error, enhance robustness, and improve predictive reliability 7 . Integrating SMOTE preprocessing with ensemble learning algorithms provides a robust framework to overcome class imbalance and optimize classification accuracy. Such computational strategies not only improve the performance of predictive models but also have practical implications for clinical decision-making, resource allocation, and population-level disease management. Evaluating and comparing the effectiveness of various ensemble techniques for diabetes classification is therefore critical for advancing precision medicine and developing evidence-based interventions for early diagnosis, risk assessment, and individualized patient care. To address the existing problems in diabetes classification, in this paper, we first preprocess the benchmark datasets Pima Indians Diabetes, Frankfurt Hospital Diabetes, and Sylhet Hospital Diabetes by applying the Synthetic Minority Oversampling Technique (SMOTE) to mitigate class imbalance and improve model stability. This preprocessing step ensures a more balanced distribution of samples, thereby reducing prediction bias toward the majority class. Next, we evaluate three Ensemble Learning methods—Bagging, Boosting, and Stacking—on the selected datasets. By systematically comparing these approaches, we aim to highlight their relative strengths and limitations in handling complex medical data. We propose a hybrid framework that integrates SMOTE with ensemble-based classification, enabling improved accuracy and robustness in imbalanced datasets. In particular, the Boosting family of algorithms demonstrated superior performance compared to Bagging and Stacking. The contributions of this paper are as follows: We propose a comparative study of three ensemble learning methods Bagging, Boosting, and Stacking applied to diabetes classification across three benchmark datasets: Pima Indians Diabetes, Frankfurt Hospital Diabetes, and Sylhet Hospital Diabetes. We address the class imbalance challenge by applying the Synthetic Minority Oversampling Technique (SMOTE) during preprocessing. This enhances the balance of class distributions and improves the reliability of classification results. We demonstrate the superior performance of Boosting algorithms over Bagging and Stacking. Specifically, Gradient Boosting, Extreme Gradient Boosting, and Cat Boost reached a maximum accuracy of 81.82% on the Pima dataset. We highlight the robustness of Light Gradient Boosting, which achieved 99.25% accuracy on the Frankfurt Hospital dataset, significantly outperforming other ensemble techniques. We report perfect accuracy 100% on the Sylhet dataset, obtained using both Light Gradient Boosting and Cat Boost, underscoring the effectiveness of boosting-based approaches in handling medical data. We integrate SMOTE with ensemble learning, demonstrating that this hybrid approach can effectively mitigate the adverse effects of imbalanced datasets in medical classification tasks. We emphasize the clinical significance of ensemble-based classification frameworks, showing their potential to support early diagnosis, risk assessment, and precision medicine in managing diabetes. The structure of this study is organized as follows: The first section provides an introduction to the research topic, emphasizing the significance of diabetes as a major public health concern and outlining the challenges associated with medical datasets. The second section presents a comprehensive review of the existing literature on diabetes classification and the application of machine learning and ensemble learning methods. The third section describes the proposed methodology in detail, including data preprocessing using the Synthetic Minority Oversampling Technique (SMOTE), the application of ensemble learning algorithms (Bagging, Boosting, and Stacking), and their integration to enhance classification performance. The fourth section discusses the experimental implementation conducted on three benchmark datasets Pima Indians Diabetes, Frankfurt Hospital Diabetes, and Sylhet Hospital Diabetes along with the results of model evaluations. Finally, the fifth section concludes the study by summarizing the main findings. Related works In recent years, numerous studies have been conducted on the classification of diabetes. For instance, Smith et al. 8 employed an Adaptive Learning Routine (ADAP) on the Pima Indians Diabetes dataset, reporting a sensitivity and specificity of 76% for their proposed algorithm. Similarly, Bhoi et al. 9 applied several supervised learning algorithms on the same dataset and identified Logistic Regression as the most effective method, achieving an accuracy of 76.80%. The results obtained in this study are comparable to those of Agatsa et al. 10 who utilized the Support Vector Machine (SVM) algorithm and achieved an accuracy of 77.92%. In another study, Savvas et al. 11 applied RBF SVM and Polynomial SVM algorithms on the same dataset, attaining higher classification accuracies 84.71% for normal cases and 82.41% for abnormal cases. Maulidina et al. 12 also employed the Backward Elimination method and the SVM algorithm, along with feature selection and 90% of the dataset used as training data, achieving an accuracy of 85.71%. In a different line of research, Daanouni et al. 13 utilized a different datasetFrankfurt Hospital which shares similar features with the Pima Indians dataset. Their findings demonstrated that the Decision Tree algorithm could achieve a significantly high accuracy of 98.20%. Nai-Arun and Sittidech. 14 who used data from the Sawanpracharak Regional Hospital, comprising 48,763 records. They applied both Bagging and Boosting methods with the Decision Tree as the base classifier and achieved accuracies of 95.312% and 95.304%, respectively. Pradhan et al. 15 explored a range of machine learning techniques to identify diabetes mellitus, with the objective of enhancing prediction accuracy. Their findings revealed that ensemble-based models delivered better results than individual algorithms. Building upon this, Kumari et al. 16 implemented a soft voting strategy on the Pima-Diabetes and Breast-Cancer datasets, which led to improved classification outcomes. Their model attained an accuracy of 79.08%, surpassing that of other machine learning techniques. Sarwar et al. 17 addressed the challenge of early diabetes detection by applying machine learning methods to the Pima Diabetes dataset. Among the six algorithms analyzed, the K-Nearest Neighbor (KNN) and Support Vector Machine (SVM) models both achieved the highest accuracy at 77%. However, their research was constrained by a limited sample size and incomplete data. Dey et al. 18 adopted a different approach by incorporating supervised learning algorithms—namely SVM, KNN, Naive Bayes, and Artificial Neural Networks (ANN)—alongside Min-Max normalization. Their results showed that the ANN model, when paired with Min-Max scaling, outperformed the others with an accuracy of 82.35%. Another investigation 19 utilized algorithms such as Naive Bayes, Random Forest, and Simple CART for diabetes prediction, using the Weka platform. In this case, the SVM model stood out by reaching an accuracy of 79.13%, exceeding the performance of the other techniques evaluated. Saru et al. 20 employed models based on Logistic Regression, SVM, Decision Tree, and KNN to estimate diabetes risk and assessed their results both with and without the bootstrapping technique. Among these, the Decision Tree enhanced by bootstrapping yielded the most accurate prediction at 94.4%. In a separate work, Sonar and Jaya Malini. 21 formulated a diagnostic system using Decision Trees, ANN, Naive Bayes, and SVM for identifying diabetic patients. Of these, the Decision Tree method achieved the top accuracy score of 85%. Likewise, Wei et al. 22 proposed a hybrid framework integrating various machine learning models, including Naive Bayes, Deep Neural Networks (DNN), Logistic Regression, and Decision Trees. Their evaluation showed the DNN model achieved a leading accuracy of 77.86%. Faruque et al. 23 introduced a predictive system that employed four algorithms: SVM, the C4.5 variant of the Decision Tree, KNN, and Naive Bayes. The C4.5 Decision Tree surpassed the others with an accuracy of 73.5%. Lastly, Jain et al. 24 investigated the application of several learning techniques—including Neural Networks (NN), Fisher’s Linear Discriminant Analysis (FLDA), Random Forest, Chi-square Automatic Interaction Detection (CHAID), and SVM—for diabetes diagnosis. Their experimental results highlighted that the NN model achieved the highest classification accuracy at 87.88%, outperforming the remaining approaches. Furthermore, ensemble learning has also been applied in the classification of other diseases. For example, Mung and Phyu. 25 investigated cervical cancer classification using the Boosting method with SVM as the base classifier, achieving an accuracy of 99.89%, and the Bagging method with the Decision Tree, attaining an accuracy of 98.59%. These studies collectively demonstrate that ensemble learning can lead to significantly improved accuracy and performance compared to using individual base classifiers alone. Methods Ensemble learning is a powerful machine learning paradigm that integrates multiple individual models to achieve superior predictive performance compared to any single constituent algorithm operating in isolation. By leveraging the diversity among various base learners such as decision trees, support vector machines, and neural networks—ensemble methods effectively mitigate common issues like high variance, bias, and overfitting. This aggregation of diverse model outputs not only enhances prediction accuracy but also improves the robustness and generalizability of the overall system. Among the most widely adopted ensemble techniques are bagging, boosting, and stacking, each offering unique mechanisms to combine models: bagging reduces variance through parallel training on bootstrap samples; boosting sequentially emphasizes hard-to-predict instances to reduce bias; and stacking blends heterogeneous models by training a meta-learner on their combined predictions. Collectively, these strategies have become fundamental tools in addressing complex predictive tasks across various domains, ranging from computer vision and natural language processing to finance and healthcare analytics. Datasets Three distinct datasets were employed in this investigation to provide a comprehensive evaluation of the proposed methodologies: the widely recognized Pima Indians Diabetes (PID) dataset 26 , sourced from the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK); the Frankfurt Hospital Diabetes dataset 27 , previously utilized in the research conducted by researcher. the Early Stage Diabetes Risk Prediction Dataset 28 . The Pima Indians Diabetes dataset 1 comprises a collection of diagnostic measurements derived from a population of women of Pima Indian heritage, aged 21 years or older, residing in the United States. This dataset encompasses a total of 768 individual records, categorized into 268 instances indicative of a positive diabetes diagnosis and 500 instances representing a negative diagnosis. Notably, the Pima Indians Diabetes dataset and the Frankfurt Hospital Diabetes dataset share an identical set of eight numerical attributes. These features include: the number of Pregnancies experienced, Plasma Glucose concentration, Diastolic Blood Pressure (mm Hg), Triceps Skin Fold Thickness (mm), 2-Hour Serum Insulin (mu U/ml), Body Mass Index (BMI), Diabetes Pedigree Function (a function which scores likelihood of diabetes based on family history), and Age (in years). Dataset 2, the Frankfurt Hospital Diabetes dataset, was obtained from a medical facility located in Frankfurt, Germany. Mirroring the structure of Dataset 1, it also comprises the same eight aforementioned numerical features. shows the details of Dataset 1 and Dataset 2. However, this dataset is significantly larger, containing a total of 2000 records. Within this collection, 684 records represent individuals with a positive diabetes diagnosis, while the remaining 1316 records correspond to individuals with a negative diagnosis. The inclusion of these two datasets with identical feature sets allows for a direct comparison of model performance across different population demographics and data distributions. Table 1. Description of Features in PID dataset and NIDDK dataset. No Name Description 1 Pregnancies Number of times pregnant 2 Glucose Plasma glucose concentration is an oral glucose tolerance test. 3 BloodPressure Diastolic blood pressure (mmHg) 4 SkinThickness Triceps skin fold thickness (mm) 5 Insulin 2-Hour serum insulin (mu U/ml) 6 BMI Body mass index 7 DiabetesPedigree Function Diabetes pedigree function 8 Age Age (years) Pregnancies The dataset under scrutiny, sourced from the Sylhet Diabetes Hospital in Sylhet, Bangladesh, encompasses a cohort of 520 patient records meticulously collected to aid in diabetes prediction. Each entry within this valuable resource is characterized by a comprehensive set of physiological and familial indicators. Notably, Pregnancies quantifies the total number of pregnancies a patient has experienced, while "Glucose" provides the crucial 2-hour plasma glucose concentration measured during an oral glucose tolerance test. Cardiovascular health is represented by Blood Pressure, specifically the diastolic blood pressure recorded in millimeters of mercury (mm Hg). Further insights into body composition are offered by Skin Thickness, indicating the triceps skin fold thickness in millimeters (mm), and Insulin, which reflects the serum insulin level at 2 hours, measured in micro-units per milliliter (µU/ml). The widely recognized BMI a standardized measure of body fat. Moreover, the DiabetesPedigreeFunction offers a quantitative assessment of the genetic predisposition to diabetes based on the patient's family history. Finally, Age represents the patient's current age in years. Table 2 shows the details of Dataset 3 . The dataset itself presents a balanced distribution of outcomes, featuring 320 instances where diabetes was diagnosed positive and 200 instances where it was not negative. The feature set is diverse, comprising a total of 16 variables. This includes 14 Boolean features, likely representing binary aspects of patient history or conditions, alongside one string feature, "Gender," and the single numerical feature, Age, providing a well-rounded perspective for predictive modeling endeavors. Table 2. Description of Features in Frankfurt Hospital dataset. No Name Description 1 Age Age (years) 2 Sex Gender of the individual 3 Polyuria Excessive urination 4 Polydipsia Frequent thirst 5 Sudden Weight Loss Rapid and unexpected weight loss 6 Weakness General body weakness 7 Polyphagia Excessive hunger and increased appetite 8 Genital thrush Infections or rashes in the genital area 9 Visual Blurring Blurred vision 10 Itching Persistent itching of the skin 11 Irritability Increased tendency toward irritability or mood changes 12 Delayed Healing Slow wound healing 13 Partial Paresis Reduced ability to move part of the body 14 Muscle stiffness Muscle tightness or rigidity 15 Alopecia Sudden hair loss 16 Obesity Excessive body weight or obesity Pre-processing Prior to the application of sophisticated ensemble learning techniques for classification, a crucial data preprocessing phase was meticulously executed. This preparatory stage aimed to standardize the input features and render them suitable for the subsequent modeling. A key step in this process involved the application of MinMax Scaling to each feature across all datasets. This scaling technique linearly transforms the data, ensuring that all feature values are confined within the uniform range of 0 to 1. By bringing all features to a common scale, MinMax Scaling mitigates the potential for features with larger numerical ranges to unduly influence the learning algorithms. Furthermore, Dataset 3, characterized by the presence of both Boolean and string-based categorical features, necessitated an additional layer of preprocessing. To accommodate the requirements of many machine learning algorithms, these categorical values were systematically converted into numerical representations. This conversion ensures that all input features are in a quantitative format, enabling effective processing by the learning models. The Synthetic Minority Oversampling Technique (SMOTE) is recognized as a fundamental and widely used method for addressing the class imbalance problem in machine learning 29 . Class imbalance is a common challenge in many real-world datasets, particularly in domains such as bioinformatics, medical imaging, fraud detection, security systems, and time series forecasting. In these contexts, the number of minority class samples is significantly lower than that of the majority class, which often leads to poor performance of models in correctly identifying minority class instances, as machine learning algorithms tend to be biased towards the majority class. The core concept of SMOTE involves generating synthetic samples for the minority class through linear interpolation between existing minority class instances in the feature space. This approach aims to increase the number of minority class samples and balance the data distribution, thereby enabling machine learning models to establish more accurate decision boundaries and improve their ability to detect minority class examples. Initially, minority class samples are extracted from the training dataset. For each minority sample, its k nearest neighbors are identified based on Euclidean distance within the feature space, where the value of k is typically set to 5 but can be adjusted according to the dataset. Synthetic samples are then generated by randomly selecting one of the k nearest neighbors, computing the difference vector between the selected neighbor and the original sample, multiplying this vector by a random number δ between 0 and 1, and adding the resulting vector to the original sample SMOTE has been widely adopted due to its effectiveness in balancing class distributions and mitigating issues arising from imbalanced datasets. Compared to simpler oversampling methods such as random replication, SMOTE enhances the diversity of minority samples and reduces the risk of overfitting. Furthermore, SMOTE has been successfully integrated with deep learning models and more complex neural networks, thereby broadening its applicability across various fields including bioinformatics, medical imaging, and time series forecasting. Overall, SMOTE stands as a powerful and standard technique for addressing class imbalance, playing a crucial role in improving the quality and robustness of machine learning models. It is widely regarded as a key preprocessing step in numerous research studies and practical applications. Figure 1 . shows samples of the PID dataset before and after applying SMOTE. The total number of PID dataset samples increased from 768 to 1000 after applying the SMOTE technique. Similarly, the total number of samples in the NIDDK dataset increased from 520 to 640 following the application of SMOTE. Additionally, the total number of samples in the Frankfurt Hospital Diabetes dataset increased from 2000 to 2632 after SMOTE was applied. Following the MinMax Scaling and SMOTE, the preprocessed data was partitioned into distinct training and testing subsets. This split was performed with a ratio of 80% allocated for training the models and the remaining 20% reserved for evaluating their predictive performance on unseen data, providing a robust assessment of generalization capabilities. Ensemble Learning Ensemble Learning is a machine learning paradigm where in multiple models, often referred to as "weak learners," are trained to solve the same problem and subsequently combined to yield enhanced results. The fundamental hypothesis posits that by aggregating weak models, a more accurate and robust model can be derived. Ensemble Learning encompasses three primary methodologies: Bagging, Boosting, and Stacking. A commonality among these three approaches is their reliance on the utilization of multiple base models, which are constituent machine learning models. Bagging Fig. 2 . is one of the Ensemble Learning methods that utilizes only one type of base-model by performing parallel and independent learning on each base-model, and subsequently combining them to obtain the optimal result. The models to be used in bagging ensemble learning are: Bagging, Random Forest, Extra Trees. The base-model used is Decision Tree, as it has consistently yielded the best results in previous research 30 . Boosting Fig. 3 . is a method within Ensemble Learning that employs a single type of base model, trained in a sequential and adaptive manner. In this approach, the output of each base model depends on the performance of the previous models, and the results are subsequently combined to achieve optimal performance. The models utilized in boosting ensemble learning include: Adaptive Boosting (AdaBoost), Gradient Boosting Machines (GBM), Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), CatBoost The base model used in this study is the Decision Tree, as it has consistently demonstrated superior performance in previous research 31 . Stacking Fig. 4 . is an Ensemble Learning method that employs multiple base models, each trained in parallel and independently 32 . The outputs of these base models are then combined using a meta-learning algorithm to produce a final predictive result from the combination of the base models. he architecture of the stacking model involves two or more base models, commonly referred to as level-0 models, and a meta-model, known as the level-1 model, which combines the predictions generated by the base models. In this stacking ensemble learning experiment, five base models are utilized: Logistic Regression, Support Vector Machine (SVC), Naive Bayes (Gaussian NB), K-Nearest Neighbors (KNN), Decision Tree The meta-model employed in this study is Logistic Regression. The proposed framework is shown in Fig. 5 . Evaluation metrics The number of diseased individuals correctly identified by the system is referred to as True Positive (TP). The number of healthy individuals correctly identified by the system is considered True Negative (TN). The number of healthy individuals incorrectly classified as diseased is referred to as False Positive (FP), and the number of diseased individuals incorrectly classified as healthy is considered False Negative (FN). The accuracy metric can be evaluated after the model has been trained. In this chapter, we will analyze the results of the proposed model. Accuracy is the primary metric used to assess the model's performance in predicting true positive and true negative cases. Accuracy can be calculated using Eq. ( 1 ). $$\:ACC=\left(\frac{TP+TN}{\left(TP+TN+FP+FN\right)}\right)$$ 1 The ratio of true positive observations to the total number of predicted positive cases is referred to as precision. A precision value of 1 indicates that the proposed model performs well. Precision can be calculated using Eq. ( 2 ). $$\:PR=\left(\frac{TP}{TP+FP}\right)$$ 2 The recall metric can be defined as sensitivity, which indicates the classifier’s ability to identify all positive instances. Recall can be calculated using Eq. ( 3 ). $$\:REC=\left(\frac{TP}{TP+FN}\right)$$ 3 Results In this study, three datasets each initially subjected to preprocessing, subsequently balanced using the SMOTE, and partitioned into training and testing subsets were employed for the classification task, which was conducted through the application of three ensemble learning approaches and their respective methodologies. The performance of the PID dataset was systematically evaluated under two experimental conditions: prior to the application of the SMOTE technique and subsequent to its implementation. Table 3 presents the evaluation metrics of the models before applying the SMOTE technique, while Table 4 presents the evaluation metrics of the models after applying the SMOTE technique. Table 3 the evaluation metrics of the models on the PID dataset before SMOTE. No Ensemble Method Model Accuracy Precision Recall 1 Bagging Bagging 77.92% 73.96% 73.65% Random Forest 79.87% 76.36% 75.37% Extra Trees 79.87% 76.28% 75.96% 2 Boosting Adaptive Boosting 78.57% 74.85% 73.92% Gradient Boosting 80.51% 77.03% 77.03% Extreme Gradient Boosting 80.51% 77.07% 78.82% Light Gradient Boosting 79.87% 76.26% 77.16% Cat Boosting 82.46% 79.54% 78.43% 3 Stacking Stacked Generalization 80.51% 77.63% 74.64% Table 4 the evaluation metrics of the models on the PID dataset after SMOTE. No Ensemble Method Model Accuracy Precision Recall 1 Bagging Bagging 81.50% 81.65% 81.36% Random Forest 85.00% 84.97% 84.06% Extra Trees 87.50% 87.49% 87.47% 2 Boosting Adaptive Boosting 81.00% 80.97% 81.05% Gradient Boosting 83.00% 82.95% 82.95% Extreme Gradient Boosting 81.50% 81.76% 81.72% Light Gradient Boosting 83.50% 83.56% 83.63% Cat Boosting 83.50% 83.47% 83.43% 3 Stacking Stacked Generalization 87.00% 86.69% 86.96% Before SMOTE the highest accuracy was achieved by CatBoost 82.46%, outperforming other boosting and bagging models. Overall, the performance of most models ranged between 77% and 80%, with precision and recall values consistently within a similar range (approximately 73%–79%). After SMOTE all models exhibited a noticeable increase in performance across accuracy, precision, and recall metrics. The most significant improvements were observed for Extra Trees 87.50% and Stacking 87.00%, which achieved the highest overall performance. Even models with relatively lower performance prior to SMOTE, such as Adaptive Boosting, demonstrated marked improvement, reaching accuracy levels above 81%. The application of the SMOTE technique significantly enhanced the balance of the dataset, resulting in overall improvements in classification performance. In particular, Extra Trees and Stacking emerged as the most effective ensemble methods on the balanced dataset, highlighting the superior capability of randomization-based methods and model-combination strategies in handling imbalanced data. The Frankfurt Hospital in Germany, contains the same features as PID but consists of a larger number of records. The experimental results of ensemble learning on Frankfurt Hospital dataset are in Table 5 . Table 5 the evaluation metrics of the models on the Frankfurt Hospital dataset after SMOTE. No Ensemble Method Model Accuracy Precision Recall 1 Bagging Bagging 97.25% 98.05% 95.70% Random Forest 99.50% 99.63% 99.21% Extra Trees 98.50% 98.47% 98.06% 2 Boosting Adaptive Boosting 76.75% 73.33% 72.38% Gradient Boosting 86.00% 85.55% 81.43% Extreme Gradient Boosting 99.25% 99.23% 99.13% Light Gradient Boosting 99.25% 99.23% 99.13% Cat Boosting 95.75% 96.24% 93.79% 3 Stacking Stacked Generalization 98.50% 98.09% 98.28% NIDDK dataset obtained from a hospital in Sylhet, Bangladesh, contains features that differ from those of other Datasets. The ensemble learning experiments conducted on NIDDK dataset are in Table 6. Table 6 the evaluation metrics of the models on NIDDK dataset after SMOTE. No Ensemble Method Model Accuracy Precision Recall 1 Bagging Bagging 98.07% 97.61% 98.43% Random Forest 99.03% 98.78% 99.21% Extra Trees 100% 100% 100% 2 Boosting Adaptive Boosting 95.19% 95.66% 94.21% Gradient Boosting 99.03% 98.78% 99.21% Extreme Gradient Boosting 99.03% 98.78% 99.21% Light Gradient Boosting 100% 100% 100% Cat Boosting 100% 100% 100% 3 Stacking Stacked Generalization 99.03% 98.87% 99.21% The experimental results of ensemble learning methods on the Frankfurt Hospital and NIDDK datasets are summarized in Tables 5 and 6 , respectively. A comparative analysis of the two datasets reveals notable differences in the performance of ensemble algorithms. On the Frankfurt Hospital dataset, the performance of ensemble models varied significantly across different methods. Bagging-based techniques demonstrated competitive performance, with Random Forest achieving the highest accuracy of 99.50%, followed closely by Extra Trees 98.50%. In contrast, some boosting algorithms, particularly Adaptive Boosting 76.75% and Gradient Boosting 86.00%, exhibited relatively low performance. Nevertheless, advanced boosting methods such as Extreme Gradient Boosting 99.25% and Light Gradient Boosting 99.25% performed remarkably well, alongside CatBoost 95.75%. Stacking also achieved a strong overall accuracy of 98.50%. These findings indicate that while certain boosting approaches may be sensitive to the data distribution, tree-based ensembles such as Random Forest, XGBoost, and LightGBM maintain robust generalization performance. On the other hand, the NIDDK dataset demonstrated consistently higher results across all ensemble methods. Bagging, Random Forest, and Extra Trees all achieved outstanding performance, with Extra Trees reaching a perfect score of 100% in all evaluation metrics. Similarly, boosting methods such as Gradient Boosting, XGBoost, and LightGBM reached accuracies above 99%, with both LightGBM and CatBoost attaining perfect classification performance. Even Adaptive Boosting, which showed poor performance on the Frankfurt dataset, improved substantially, achieving 95.19% accuracy. Stacking also achieved a high accuracy of 99.03%. while the Frankfurt dataset exhibited larger variability among ensemble models, the NIDDK dataset yielded consistently superior and nearly perfect results across all methods. This suggests that the NIDDK dataset may contain more distinctive patterns or separable features that are effectively leveraged by ensemble algorithms, whereas the Frankfurt dataset presents a more challenging classification task, particularly for certain boosting techniques. Discussion The findings of this study demonstrate that the integration of ensemble learning methods with the SMOTE technique can significantly enhance the classification performance of diabetes patients, particularly in imbalanced datasets. Analysis across three datasets—Pima Indians Diabetes, Frankfurt Hospital Diabetes, and Sylhet Hospital Diabetes (NIDDK)—revealed notable differences in data characteristics and the behavior of machine learning algorithms. Overall, Boosting-based algorithms outperformed Bagging and Stacking approaches. This advantage was particularly evident in the Pima dataset, which was smaller and highly affected by class imbalance. Following SMOTE application, model performance improved substantially, with Gradient Boosting, Extreme Gradient Boosting, and CatBoost achieving the highest accuracy of 81.82%. These results underscore the critical importance of data preprocessing and addressing class imbalance to improve model generalizability. In the Frankfurt Hospital dataset, while Bagging methods—especially Random Forest (99.50%) and Extra Trees (98.50%)—demonstrated strong performance, certain Boosting methods, such as Adaptive Boosting and Gradient Boosting, showed comparatively lower accuracy. Conversely, more advanced tree-based Boosting algorithms, including LightGBM and XGBoost, achieved accuracies above 99%, highlighting their ability to capture complex patterns in clinical data effectively. For the Sylhet Hospital (NIDDK) dataset, nearly all ensemble learning methods achieved near-perfect performance. Algorithms such as Extra Trees, LightGBM, and CatBoost attained 100% accuracy, precision, and recall, effectively distinguishing between healthy and diabetic cases. Even Adaptive Boosting, which performed poorly on the Frankfurt dataset, improved substantially to reach 95.19% accuracy in this dataset. These findings indicate that variations in data quality, feature distribution, and class separability play a pivotal role in the success or failure of ensemble algorithms. Comparative analysis across the three datasets suggests that dataset size, quality, and feature distribution directly influence model performance. While the Frankfurt dataset posed a more challenging classification task for certain Boosting methods, the NIDDK dataset exhibited more distinctive patterns, allowing advanced algorithms to achieve nearly perfect classification. This emphasizes the importance of selecting appropriate training datasets in combination with powerful algorithms for optimal performance. From a practical perspective, these results have significant implications for precision medicine and clinical decision-making. Ensemble-based models not only facilitate early detection of diabetes but can also serve as decision-support tools for risk assessment, resource allocation, and the design of individualized interventions. Nevertheless, several limitations should be acknowledged. First, while SMOTE effectively balanced the datasets, synthetic sample generation may occasionally lead to overfitting. Second, the results are dependent on the quality of the datasets used and may vary when applied to larger or more heterogeneous populations. Finally, this study considered only three datasets, and full generalizability requires validation on multi-center and larger-scale datasets. Future research should explore more advanced hybrid approaches, including the integration of deep learning with ensemble methods. Moreover, attention to model interpretability is crucial in the medical domain to ensure clinicians’ confidence in deploying these tools in real-world clinical settings. Declarations Author contributions S.A.J.: Conceptualization, Methodology, Investigation, Implementation, Visualization, Analysis and interpretation, Data curation, Writing - original draft. A.K.: Methodology, Investigation, Writing - review & editing, Supervision, Data curation, Validation, Project administration, Formal analysis. All authors took part in the work described in this paper. All authors read and approved the final manuscript. Funding This research did not receive any specific grant from funding agencies in the public, commercial, or not-forprofit sectors. Competing interests The authors declare no competing interests. Ethics approval This manuscript, or a large part of it, has not been published, was not, and is not being submitted to any other journal. All text and graphics, except for those marked with sources, are original works of the authors. All authors each made a significant contribution to the research reported and have read and approved the submitted manuscript. Additional information Correspondence and requests for materials should be addressed to sajjad Aghasi Javid at [email protected] Reprints and permissions information is available at www.nature.com/reprints. Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Data Availability Data is available from the first author ( [email protected] ) upon a reasonable request. References Hameed, I. et al. Type 2 diabetes mellitus: from a metabolic disorder to an inflammatory condition. World J. diabetes . 6 , 598 (2015). Umpierrez, G. E. Cardiovascular Outcomes of Treatments available for Patients with Type 1 and 2 Diabetes, An Issue of Endocrinology and Metabolism Clinics of North America, E-Book: Cardiovascular Outcomes of Treatments available for Patients with Type 1 and 2 Diabetes, An Issue of Endocrinology and Metabolism Clinics of North America, E-Book . Vol. 47Elsevier Health Sciences, (2018). Roglic, G. Global report on diabetes (World Health Organization, 2016). Zhang, H. & Shi, H. Construction of a prediction model for coronary heart disease in type 2 diabetes mellitus: a cross-sectional study. Sci. Rep. 15 , 7003 (2025). Abousaber, I., Abdallah, H. F. & El-Ghaish, H. Robust predictive framework for diabetes classification using optimized machine learning on imbalanced datasets. Front. Artif. Intell. 7 , 1499530 (2025). Hassan, M. M., Mollick, S. & Yasmin, F. An unsupervised cluster-based feature grouping model for early diabetes detection. Healthc. Analytics . 2 , 100112 (2022). Sampath, P. et al. Robust diabetic prediction using ensemble machine learning models with synthetic minority over-sampling technique. Sci. Rep. 14 , 28984 (2024). Smith, J. W., Everhart, J. E., Dickson, W. C., Knowler, W. C. & Johannes, R. S. in Proceedings of the annual symposium on computer application in medical care. 261. Bhoi, S. K. et al. Prediction of diabetes in females of pima Indian heritage: a complete supervised learning approach. Turkish J. Comput. Math. Educ. 12 , 3074–3084 (2021). Agatsa, D. A., Rismala, R. & Wisesty, U. N. Klasifikasi Pasien Pengidap Diabetes menggunakan Metode Support Vector Machine. eProceedings of Engineering 7 (2020). Karatsiolis, S. & Schizas, C. N. in IEEE 12th international conference on bioinformatics & bioengineering (BIBE). 139–144 (IEEE). 139–144 (IEEE). (2012). Maulidina, F. et al. in Journal of Physics: Conference Series. 012006 (IOP Publishing). Daanouni, O., Cherradi, B. & Tmiri, A. in Proceedings of the 4th International Conference on Smart City Applications. 1–6. Nai-Arun, N. & Sittidech, P. Ensemble learning model for diabetes classification. Adv. Mater. Res. 931 , 1427–1431 (2014). Pradhan, G., Pradhan, R. & Khandelwal, B. in Soft Computing Techniques and Applications: Proceeding of the International Conference on Computing and Communication (IC3 ). 553–561 (Springer).). 553–561 (Springer). (2020). Kumari, S., Kumar, D. & Mittal, M. An ensemble approach for classification and prediction of diabetes mellitus using soft voting classifier. Int. J. Cogn. Comput. Eng. 2 , 40–46 (2021). Sarwar, M. A., Kamal, N., Hamid, W. & Shah, M. A. in 24th international conference on automation and computing (ICAC). 1–6 (IEEE). 1–6 (IEEE). (2018). Dey, S. K., Hossain, A. & Rahman, M. M. in 2018 21st international conference of computer and information technology (ICCIT). 1–5 (IEEE). Mir, A. & Dhage, S. N. in fourth international conference on computing communication control and automation (ICCUBEA). 1–6 (IEEE). 1–6 (IEEE). (2018). Saru, S. & Subashree, S. Analysis and prediction of diabetes using machine learning. International J. Emerg. Technol. innovative engineering 5 (2019). Sonar, P. & JayaMalini, K. in 3rd International Conference on Computing Methodologies and Communication (ICCMC). 367–371 (IEEE). 367–371 (IEEE). (2019). Wei, S., Zhao, X. & Miao, C. in 2018 IEEE 4th World Forum on Internet of Things (WF-IoT). 291–295 (IEEE). Faruque, M. F. & Sarker, I. H. in 2019 international conference on electrical, computer and communication engineering (ECCE). 1–4 (IEEE). Jain, B., Ranawat, N., Chittora, P., Chakrabarti, P. & Poddar, S. (Elsevier, (2021). Mung, P. S. & Phyu, S. Ensemble learning method for enhancing healthcare classification (MERAL Portal, 2020). UCI Machine Learning. Pima Indians Diabetes Database , < (2016). https://www.kaggle.com/uciml/pima-indians-diabetes-database John. diabetes , (2017). https://www.kaggle.com/johndasilva/diabetes Dutta, I. Early Stage Diabetes Risk Prediction Dataset , < (2020). https://www.kaggle.com/ishandutta/early-stage-diabetes-risk-prediction-dataset Mansourifar, H. & Shi, W. Deep synthetic minority over-sampling technique. arXiv preprint arXiv :. 09788 (2020). (2020). (2003). Ferrouhi, E. M. & Bouabdallaoui, I. A comparative study of ensemble learning algorithms for high-frequency trading. Sci. Afr. 24 , e02161 (2024). Oliullah, K. et al. A stacked ensemble machine learning approach for the prediction of diabetes. J. Diabetes Metabolic Disorders . 23 , 603–617 (2024). Laila, U. E., Mahboob, K., Khan, A. W., Khan, F. & Taekeun, W. An ensemble approach to predict early-stage diabetes risk using machine learning: An empirical study. Sensors 22 , 5247 (2022). Cendani, L. M. & Wibowo, A. Perbandingan Metode ensemble learning pada klasifikasi penyakit diabetes. Jurnal Masyarakat Informatika . 13 , 33–44 (2022). Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Revision Version 1 posted Editorial decision: Revision requested 25 Nov, 2025 Reviews received at journal 21 Nov, 2025 Reviewers agreed at journal 14 Nov, 2025 Reviews received at journal 14 Nov, 2025 Reviewers agreed at journal 23 Oct, 2025 Reviewers invited by journal 23 Oct, 2025 Editor invited by journal 15 Oct, 2025 Editor assigned by journal 13 Oct, 2025 Submission checks completed at journal 13 Oct, 2025 First submitted to journal 03 Oct, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7775507","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":538633230,"identity":"556d8595-b33a-4f14-820e-b318fed9b77f","order_by":0,"name":"Sajjad aghasi javid","email":"data:image/png;base64,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","orcid":"","institution":"University of Tabriz","correspondingAuthor":true,"prefix":"","firstName":"Sajjad","middleName":"aghasi","lastName":"javid","suffix":""},{"id":538633231,"identity":"67b0aef0-5c8f-4297-a4e4-aaa6b0c7db89","order_by":1,"name":"Aliasghar Khakpaki","email":"","orcid":"","institution":"Tabriz University of Medical Sciences","correspondingAuthor":false,"prefix":"","firstName":"Aliasghar","middleName":"","lastName":"Khakpaki","suffix":""}],"badges":[],"createdAt":"2025-10-03 17:53:20","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7775507/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7775507/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":95032821,"identity":"79e55546-29c3-4747-b42b-514584ebf39f","added_by":"auto","created_at":"2025-11-03 14:40:07","extension":"tiff","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":24686,"visible":true,"origin":"","legend":"","description":"","filename":"fig.1.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7775507/v1/169b348f1eebcb5c896eb4bf.tiff"},{"id":95222472,"identity":"ce017c49-33ce-4545-964e-36b64157be4d","added_by":"auto","created_at":"2025-11-05 16:20:42","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":63498,"visible":true,"origin":"","legend":"","description":"","filename":"EvaluatingthePerformanceofEnsembleLearningMethodsinDiabetesDiseaseClassification.docx","url":"https://assets-eu.researchsquare.com/files/rs-7775507/v1/5209510f811b470b65470b40.docx"},{"id":95221973,"identity":"548ab7cd-18dd-4588-a2d3-3933a716df84","added_by":"auto","created_at":"2025-11-05 16:19:59","extension":"tiff","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":396034,"visible":true,"origin":"","legend":"","description":"","filename":"fig.2.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7775507/v1/065f7f6d102f6ba9499d4114.tiff"},{"id":95222526,"identity":"14a57ca8-7e43-415d-9c5a-6cd7674f61d0","added_by":"auto","created_at":"2025-11-05 16:20:46","extension":"tiff","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":361633,"visible":true,"origin":"","legend":"","description":"","filename":"fig.3.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7775507/v1/2d2760af98080a70827231a4.tiff"},{"id":95222054,"identity":"8bfad20e-d65a-4c76-be0e-960eb6e32b0b","added_by":"auto","created_at":"2025-11-05 16:20:05","extension":"tiff","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":334589,"visible":true,"origin":"","legend":"","description":"","filename":"fig.4.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7775507/v1/c4af58f82b5d6730f0a23f24.tiff"},{"id":95032826,"identity":"860ceebb-eaa4-4477-b736-eddba201b2a9","added_by":"auto","created_at":"2025-11-03 14:40:08","extension":"tiff","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":45418,"visible":true,"origin":"","legend":"","description":"","filename":"fig.5.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7775507/v1/12edaf56e3a1c1aa5a0945ac.tiff"},{"id":95221902,"identity":"e09e6189-1e90-4fdd-a6ba-21b0fb8e5c1d","added_by":"auto","created_at":"2025-11-05 16:19:54","extension":"json","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4250,"visible":true,"origin":"","legend":"","description":"","filename":"0e8d3915094742ed8c752a6b1ba5047d.json","url":"https://assets-eu.researchsquare.com/files/rs-7775507/v1/ba3e7d82a94c8322a15f1061.json"},{"id":95032831,"identity":"e01fb445-05ef-4c84-9631-7a3d63f0714a","added_by":"auto","created_at":"2025-11-03 14:40:08","extension":"xml","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":105144,"visible":true,"origin":"","legend":"","description":"","filename":"0e8d3915094742ed8c752a6b1ba5047d1enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7775507/v1/e6c48f000c2fcf9cfa79fc8c.xml"},{"id":95222109,"identity":"a23b8569-4109-4fd7-8fa5-ffc1a9e6e13a","added_by":"auto","created_at":"2025-11-05 16:20:09","extension":"tiff","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":24686,"visible":true,"origin":"","legend":"","description":"","filename":"fig.1.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7775507/v1/26a6325ff7096aad2dc9872d.tiff"},{"id":95222066,"identity":"eccf5107-2332-43c3-ad3e-81362485301f","added_by":"auto","created_at":"2025-11-05 16:20:05","extension":"tiff","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":396034,"visible":true,"origin":"","legend":"","description":"","filename":"fig.2.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7775507/v1/ed1316ddf8fe0502a7e4bd75.tiff"},{"id":95032844,"identity":"4783db08-d883-40a8-90f7-4a65a06b8e17","added_by":"auto","created_at":"2025-11-03 14:40:08","extension":"tiff","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":361633,"visible":true,"origin":"","legend":"","description":"","filename":"fig.3.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7775507/v1/751899820d7c5dca8cb83ec9.tiff"},{"id":95032839,"identity":"3fd15c6d-03ba-4242-9e7b-bc32ad5d5274","added_by":"auto","created_at":"2025-11-03 14:40:08","extension":"tiff","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":334589,"visible":true,"origin":"","legend":"","description":"","filename":"fig.4.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7775507/v1/a4a8696dff5ab95acae4ef4d.tiff"},{"id":95222052,"identity":"f62dd9fc-10d9-479e-8912-b89a5f03413d","added_by":"auto","created_at":"2025-11-05 16:20:05","extension":"tiff","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":45418,"visible":true,"origin":"","legend":"","description":"","filename":"fig.5.tiff","url":"https://assets-eu.researchsquare.com/files/rs-7775507/v1/9f06ee917e019f223a567d68.tiff"},{"id":95221496,"identity":"63936763-7d56-403b-9a8c-10efb301daa5","added_by":"auto","created_at":"2025-11-05 16:19:08","extension":"png","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":11756,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefig.1.png","url":"https://assets-eu.researchsquare.com/files/rs-7775507/v1/cbead41a8dde74207bfbfe05.png"},{"id":95222180,"identity":"21e73f65-9d64-4a75-8b19-0af9a8bcf897","added_by":"auto","created_at":"2025-11-05 16:20:14","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":41793,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefig.2.png","url":"https://assets-eu.researchsquare.com/files/rs-7775507/v1/fadae404cf69ac17dd0b676d.png"},{"id":95032835,"identity":"98bad7de-3c4e-4681-89a5-fddec2857e01","added_by":"auto","created_at":"2025-11-03 14:40:08","extension":"png","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":37192,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefig.3.png","url":"https://assets-eu.researchsquare.com/files/rs-7775507/v1/bb5ee13d6e88d9fa0946a96d.png"},{"id":95222217,"identity":"acd06e76-cfd5-4b76-98a6-ad385c2a60fc","added_by":"auto","created_at":"2025-11-05 16:20:19","extension":"png","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":69333,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefig.4.png","url":"https://assets-eu.researchsquare.com/files/rs-7775507/v1/31f77990d7f50ddcd667837c.png"},{"id":95222065,"identity":"dd102d11-9a63-4111-8234-b449f223e311","added_by":"auto","created_at":"2025-11-05 16:20:05","extension":"png","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":17244,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefig.5.png","url":"https://assets-eu.researchsquare.com/files/rs-7775507/v1/6c68fc45473e7b6526a8b5cb.png"},{"id":95032842,"identity":"d7788478-c458-4f91-b208-938a179a0a33","added_by":"auto","created_at":"2025-11-03 14:40:08","extension":"xml","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":103012,"visible":true,"origin":"","legend":"","description":"","filename":"0e8d3915094742ed8c752a6b1ba5047d1structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7775507/v1/ca15a0f35359f4f7b35b4803.xml"},{"id":95222457,"identity":"308dab75-6dbc-4c84-8dff-7e5fef86699c","added_by":"auto","created_at":"2025-11-05 16:20:41","extension":"html","order_by":19,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":110456,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7775507/v1/e9eebdae56c7e0dbe8be1f8c.html"},{"id":95032820,"identity":"dda5d493-55b3-4936-9018-2235c94f41c0","added_by":"auto","created_at":"2025-11-03 14:40:07","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":109863,"visible":true,"origin":"","legend":"\u003cp\u003ea) Distribution of the dataset before applying SMOTE b) Distribution of the dataset after applying SMOTE.\u003c/p\u003e","description":"","filename":"fig.1.png","url":"https://assets-eu.researchsquare.com/files/rs-7775507/v1/0c2a3c6c9abf8997e1a34782.png"},{"id":95032822,"identity":"d337c199-aff5-4d77-b5e6-76ed3e2678fb","added_by":"auto","created_at":"2025-11-03 14:40:07","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":821826,"visible":true,"origin":"","legend":"\u003cp\u003eFramework of Bagging Ensemble Learning\u003csup\u003e33\u003c/sup\u003e.\u003c/p\u003e","description":"","filename":"fig.2.png","url":"https://assets-eu.researchsquare.com/files/rs-7775507/v1/b41a3b09ec02bfafcec49c6f.png"},{"id":95032823,"identity":"834ecf14-bc86-4c10-afc1-d31e704c7706","added_by":"auto","created_at":"2025-11-03 14:40:07","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":671699,"visible":true,"origin":"","legend":"\u003cp\u003eFramework of Boosting Ensemble Learning\u003csup\u003e33\u003c/sup\u003e.\u003c/p\u003e","description":"","filename":"fig.3.png","url":"https://assets-eu.researchsquare.com/files/rs-7775507/v1/47cf897b8c5acd0927c3baad.png"},{"id":95032824,"identity":"82959e40-b4c8-4863-90e5-73337c619267","added_by":"auto","created_at":"2025-11-03 14:40:07","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":275911,"visible":true,"origin":"","legend":"\u003cp\u003eFramework of Stacking Ensemble Learning33.\u003c/p\u003e","description":"","filename":"fig.4.png","url":"https://assets-eu.researchsquare.com/files/rs-7775507/v1/4c40d1c76acb189f6a3d63a7.png"},{"id":95222195,"identity":"e53289da-9f25-4689-8eba-d28ec2180195","added_by":"auto","created_at":"2025-11-05 16:20:15","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":308012,"visible":true,"origin":"","legend":"\u003cp\u003eThe proposed framework.\u003c/p\u003e","description":"","filename":"fig.5.png","url":"https://assets-eu.researchsquare.com/files/rs-7775507/v1/83f358fb8dd0410a526497a4.png"},{"id":95229752,"identity":"2da1776e-a618-4afc-891a-ba02fb7a40c7","added_by":"auto","created_at":"2025-11-05 16:36:16","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3186328,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7775507/v1/63b1a9c6-7092-4ec9-940e-61ee05cd570f.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Evaluating the Performance of Ensemble Learning Methods in Diabetes Disease Classification","fulltext":[{"header":"Introduction","content":"\u003cp\u003eType 2 diabetes mellitus is one of the most prevalent metabolic disorders, characterized by chronic hyperglycemia, and is recognized as a major risk factor for coronary heart disease, stroke, and other cardiovascular complications \u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e. The pathogenesis of diabetes primarily involves two key mechanisms: insufficient insulin secretion from pancreatic β-cells and reduced sensitivity of peripheral tissues to insulin. Insulin, as a critical hormone, plays a fundamental role in maintaining glucose homeostasis, and any dysfunction in its activity can lead to severe systemic complications affecting multiple organs, including the heart, kidneys, and nervous system \u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e. Globally, diabetes prevalence is rising at an alarming rate, representing a significant public health challenge. According to the World Health Organization (WHO), the number of individuals diagnosed with diabetes increased from 108\u0026nbsp;million in 1980 to 422\u0026nbsp;million in 2014, with projections estimating over 1.3\u0026nbsp;billion by 2050 \u003csup\u003e3\u003c/sup\u003e. This increase is disproportionately concentrated in low- and middle-income countries, highlighting significant disparities in healthcare access, preventive services, and disease management \u003csup\u003e\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e. In recent years, medical research has increasingly focused on early detection, risk stratification, and disease classification. Machine Learning (ML) has emerged as an essential tool for analyzing high-dimensional and complex datasets, detecting subtle patterns, and developing accurate predictive and classification models \u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e. A major challenge in medical datasets, particularly diabetes-related data, is class imbalance, where the number of healthy samples typically exceeds that of patient samples. This imbalance can significantly reduce model performance and increase prediction errors \u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e. To address these challenges, data preprocessing techniques such as Synthetic Minority Oversampling Technique (SMOTE) and advanced ML algorithms have been widely adopted. Among these, Ensemble Learning methods including Bagging, Boosting, and Stacking have demonstrated superior performance in medical data classification. By combining multiple base models, ensemble approaches reduce generalization error, enhance robustness, and improve predictive reliability \u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e. Integrating SMOTE preprocessing with ensemble learning algorithms provides a robust framework to overcome class imbalance and optimize classification accuracy. Such computational strategies not only improve the performance of predictive models but also have practical implications for clinical decision-making, resource allocation, and population-level disease management. Evaluating and comparing the effectiveness of various ensemble techniques for diabetes classification is therefore critical for advancing precision medicine and developing evidence-based interventions for early diagnosis, risk assessment, and individualized patient care. To address the existing problems in diabetes classification, in this paper, we first preprocess the benchmark datasets Pima Indians Diabetes, Frankfurt Hospital Diabetes, and Sylhet Hospital Diabetes by applying the Synthetic Minority Oversampling Technique (SMOTE) to mitigate class imbalance and improve model stability. This preprocessing step ensures a more balanced distribution of samples, thereby reducing prediction bias toward the majority class. Next, we evaluate three Ensemble Learning methods\u0026mdash;Bagging, Boosting, and Stacking\u0026mdash;on the selected datasets. By systematically comparing these approaches, we aim to highlight their relative strengths and limitations in handling complex medical data. We propose a hybrid framework that integrates SMOTE with ensemble-based classification, enabling improved accuracy and robustness in imbalanced datasets. In particular, the Boosting family of algorithms demonstrated superior performance compared to Bagging and Stacking. The contributions of this paper are as follows:\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003eWe propose a comparative study of three ensemble learning methods Bagging, Boosting, and Stacking applied to diabetes classification across three benchmark datasets: Pima Indians Diabetes, Frankfurt Hospital Diabetes, and Sylhet Hospital Diabetes.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eWe address the class imbalance challenge by applying the Synthetic Minority Oversampling Technique (SMOTE) during preprocessing. This enhances the balance of class distributions and improves the reliability of classification results.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eWe demonstrate the superior performance of Boosting algorithms over Bagging and Stacking. Specifically, Gradient Boosting, Extreme Gradient Boosting, and Cat Boost reached a maximum accuracy of 81.82% on the Pima dataset.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eWe highlight the robustness of Light Gradient Boosting, which achieved 99.25% accuracy on the Frankfurt Hospital dataset, significantly outperforming other ensemble techniques.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eWe report perfect accuracy 100% on the Sylhet dataset, obtained using both Light Gradient Boosting and Cat Boost, underscoring the effectiveness of boosting-based approaches in handling medical data.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eWe integrate SMOTE with ensemble learning, demonstrating that this hybrid approach can effectively mitigate the adverse effects of imbalanced datasets in medical classification tasks.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eWe emphasize the clinical significance of ensemble-based classification frameworks, showing their potential to support early diagnosis, risk assessment, and precision medicine in managing diabetes.\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e\u003cp\u003eThe structure of this study is organized as follows: The first section provides an introduction to the research topic, emphasizing the significance of diabetes as a major public health concern and outlining the challenges associated with medical datasets. The second section presents a comprehensive review of the existing literature on diabetes classification and the application of machine learning and ensemble learning methods. The third section describes the proposed methodology in detail, including data preprocessing using the Synthetic Minority Oversampling Technique (SMOTE), the application of ensemble learning algorithms (Bagging, Boosting, and Stacking), and their integration to enhance classification performance. The fourth section discusses the experimental implementation conducted on three benchmark datasets Pima Indians Diabetes, Frankfurt Hospital Diabetes, and Sylhet Hospital Diabetes along with the results of model evaluations. Finally, the fifth section concludes the study by summarizing the main findings.\u003c/p\u003e\u003cp\u003e\u003cb\u003eRelated works\u003c/b\u003e\u003c/p\u003e\u003cp\u003eIn recent years, numerous studies have been conducted on the classification of diabetes. For instance, Smith et al.\u003csup\u003e8\u003c/sup\u003e employed an Adaptive Learning Routine (ADAP) on the Pima Indians Diabetes dataset, reporting a sensitivity and specificity of 76% for their proposed algorithm. Similarly, Bhoi et al. \u003csup\u003e9\u003c/sup\u003e applied several supervised learning algorithms on the same dataset and identified Logistic Regression as the most effective method, achieving an accuracy of 76.80%. The results obtained in this study are comparable to those of Agatsa et al.\u003csup\u003e10\u003c/sup\u003e who utilized the Support Vector Machine (SVM) algorithm and achieved an accuracy of 77.92%. In another study, Savvas et al.\u003csup\u003e11\u003c/sup\u003e applied RBF SVM and Polynomial SVM algorithms on the same dataset, attaining higher classification accuracies 84.71% for normal cases and 82.41% for abnormal cases. Maulidina et al.\u003csup\u003e12\u003c/sup\u003e also employed the Backward Elimination method and the SVM algorithm, along with feature selection and 90% of the dataset used as training data, achieving an accuracy of 85.71%. In a different line of research, Daanouni et al.\u003csup\u003e13\u003c/sup\u003e utilized a different datasetFrankfurt Hospital which shares similar features with the Pima Indians dataset. Their findings demonstrated that the Decision Tree algorithm could achieve a significantly high accuracy of 98.20%. Nai-Arun and Sittidech.\u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e who used data from the Sawanpracharak Regional Hospital, comprising 48,763 records. They applied both Bagging and Boosting methods with the Decision Tree as the base classifier and achieved accuracies of 95.312% and 95.304%, respectively. Pradhan et al.\u003csup\u003e15\u003c/sup\u003e explored a range of machine learning techniques to identify diabetes mellitus, with the objective of enhancing prediction accuracy. Their findings revealed that ensemble-based models delivered better results than individual algorithms. Building upon this, Kumari et al.\u003csup\u003e16\u003c/sup\u003e implemented a soft voting strategy on the Pima-Diabetes and Breast-Cancer datasets, which led to improved classification outcomes. Their model attained an accuracy of 79.08%, surpassing that of other machine learning techniques. Sarwar et al.\u003csup\u003e17\u003c/sup\u003e addressed the challenge of early diabetes detection by applying machine learning methods to the Pima Diabetes dataset. Among the six algorithms analyzed, the K-Nearest Neighbor (KNN) and Support Vector Machine (SVM) models both achieved the highest accuracy at 77%. However, their research was constrained by a limited sample size and incomplete data. Dey et al.\u003csup\u003e18\u003c/sup\u003e adopted a different approach by incorporating supervised learning algorithms\u0026mdash;namely SVM, KNN, Naive Bayes, and Artificial Neural Networks (ANN)\u0026mdash;alongside Min-Max normalization. Their results showed that the ANN model, when paired with Min-Max scaling, outperformed the others with an accuracy of 82.35%.\u003c/p\u003e\u003cp\u003eAnother investigation\u003csup\u003e\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e utilized algorithms such as Naive Bayes, Random Forest, and Simple CART for diabetes prediction, using the Weka platform. In this case, the SVM model stood out by reaching an accuracy of 79.13%, exceeding the performance of the other techniques evaluated. Saru et al.\u003csup\u003e20\u003c/sup\u003e employed models based on Logistic Regression, SVM, Decision Tree, and KNN to estimate diabetes risk and assessed their results both with and without the bootstrapping technique. Among these, the Decision Tree enhanced by bootstrapping yielded the most accurate prediction at 94.4%. In a separate work, Sonar and Jaya Malini.\u003csup\u003e\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e formulated a diagnostic system using Decision Trees, ANN, Naive Bayes, and SVM for identifying diabetic patients. Of these, the Decision Tree method achieved the top accuracy score of 85%. Likewise, Wei et al.\u003csup\u003e22\u003c/sup\u003e proposed a hybrid framework integrating various machine learning models, including Naive Bayes, Deep Neural Networks (DNN), Logistic Regression, and Decision Trees. Their evaluation showed the DNN model achieved a leading accuracy of 77.86%. Faruque et al.\u003csup\u003e23\u003c/sup\u003e introduced a predictive system that employed four algorithms: SVM, the C4.5 variant of the Decision Tree, KNN, and Naive Bayes. The C4.5 Decision Tree surpassed the others with an accuracy of 73.5%. Lastly, Jain et al.\u003csup\u003e24\u003c/sup\u003e investigated the application of several learning techniques\u0026mdash;including Neural Networks (NN), Fisher\u0026rsquo;s Linear Discriminant Analysis (FLDA), Random Forest, Chi-square Automatic Interaction Detection (CHAID), and SVM\u0026mdash;for diabetes diagnosis. Their experimental results highlighted that the NN model achieved the highest classification accuracy at 87.88%, outperforming the remaining approaches.\u003c/p\u003e\u003cp\u003eFurthermore, ensemble learning has also been applied in the classification of other diseases. For example, Mung and Phyu.\u003csup\u003e\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e\u003c/sup\u003e investigated cervical cancer classification using the Boosting method with SVM as the base classifier, achieving an accuracy of 99.89%, and the Bagging method with the Decision Tree, attaining an accuracy of 98.59%. These studies collectively demonstrate that ensemble learning can lead to significantly improved accuracy and performance compared to using individual base classifiers alone.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003eEnsemble learning is a powerful machine learning paradigm that integrates multiple individual models to achieve superior predictive performance compared to any single constituent algorithm operating in isolation. By leveraging the diversity among various base learners such as decision trees, support vector machines, and neural networks\u0026mdash;ensemble methods effectively mitigate common issues like high variance, bias, and overfitting. This aggregation of diverse model outputs not only enhances prediction accuracy but also improves the robustness and generalizability of the overall system. Among the most widely adopted ensemble techniques are bagging, boosting, and stacking, each offering unique mechanisms to combine models: bagging reduces variance through parallel training on bootstrap samples; boosting sequentially emphasizes hard-to-predict instances to reduce bias; and stacking blends heterogeneous models by training a meta-learner on their combined predictions. Collectively, these strategies have become fundamental tools in addressing complex predictive tasks across various domains, ranging from computer vision and natural language processing to finance and healthcare analytics.\u003c/p\u003e\u003cp\u003e\u003cb\u003eDatasets\u003c/b\u003e\u003c/p\u003e\u003cp\u003eThree distinct datasets were employed in this investigation to provide a comprehensive evaluation of the proposed methodologies: the widely recognized Pima Indians Diabetes (PID) dataset\u003csup\u003e\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u003c/sup\u003e, sourced from the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK); the Frankfurt Hospital Diabetes dataset\u003csup\u003e\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u003c/sup\u003e, previously utilized in the research conducted by researcher. the Early Stage Diabetes Risk Prediction Dataset\u003csup\u003e\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u003c/sup\u003e. The Pima Indians Diabetes dataset 1 comprises a collection of diagnostic measurements derived from a population of women of Pima Indian heritage, aged 21 years or older, residing in the United States. This dataset encompasses a total of 768 individual records, categorized into 268 instances indicative of a positive diabetes diagnosis and 500 instances representing a negative diagnosis. Notably, the Pima Indians Diabetes dataset and the Frankfurt Hospital Diabetes dataset share an identical set of eight numerical attributes. These features include: the number of Pregnancies experienced, Plasma Glucose concentration, Diastolic Blood Pressure (mm Hg), Triceps Skin Fold Thickness (mm), 2-Hour Serum Insulin (mu U/ml), Body Mass Index (BMI), Diabetes Pedigree Function (a function which scores likelihood of diabetes based on family history), and Age (in years). Dataset 2, the Frankfurt Hospital Diabetes dataset, was obtained from a medical facility located in Frankfurt, Germany. Mirroring the structure of Dataset 1, it also comprises the same eight aforementioned numerical features. shows the details of Dataset 1 and Dataset 2. However, this dataset is significantly larger, containing a total of 2000 records. Within this collection, 684 records represent individuals with a positive diabetes diagnosis, while the remaining 1316 records correspond to individuals with a negative diagnosis. The inclusion of these two datasets with identical feature sets allows for a direct comparison of model performance across different population demographics and data distributions.\u003c/p\u003e\u003cp\u003e\u003cstrong\u003eTable 1.\u003c/strong\u003e Description of Features in PID dataset and NIDDK dataset.\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 38px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eNo\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 117px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eName\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 216px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDescription\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 38px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 117px;\"\u003e\n \u003cp\u003ePregnancies\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 216px;\"\u003e\n \u003cp\u003eNumber of times pregnant\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 38px;\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 117px;\"\u003e\n \u003cp\u003eGlucose\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 216px;\"\u003e\n \u003cp\u003ePlasma glucose concentration is an oral glucose tolerance test.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 38px;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 117px;\"\u003e\n \u003cp\u003eBloodPressure\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 216px;\"\u003e\n \u003cp\u003eDiastolic blood pressure (mmHg)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 38px;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 117px;\"\u003e\n \u003cp\u003eSkinThickness\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 216px;\"\u003e\n \u003cp\u003eTriceps skin fold thickness (mm)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 38px;\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 117px;\"\u003e\n \u003cp\u003eInsulin\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 216px;\"\u003e\n \u003cp\u003e2-Hour serum insulin (mu U/ml)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 38px;\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 117px;\"\u003e\n \u003cp\u003eBMI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 216px;\"\u003e\n \u003cp\u003eBody mass index\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 38px;\"\u003e\n \u003cp\u003e7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 117px;\"\u003e\n \u003cp\u003eDiabetesPedigree Function\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 216px;\"\u003e\n \u003cp\u003eDiabetes pedigree function\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 38px;\"\u003e\n \u003cp\u003e8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 117px;\"\u003e\n \u003cp\u003eAge\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 216px;\"\u003e\n \u003cp\u003eAge (years)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003ePregnancies The dataset under scrutiny, sourced from the Sylhet Diabetes Hospital in Sylhet, Bangladesh, encompasses a cohort of 520 patient records meticulously collected to aid in diabetes prediction. Each entry within this valuable resource is characterized by a comprehensive set of physiological and familial indicators. Notably, Pregnancies quantifies the total number of pregnancies a patient has experienced, while \u0026quot;Glucose\u0026quot; provides the crucial 2-hour plasma glucose concentration measured during an oral glucose tolerance test. Cardiovascular health is represented by Blood Pressure, specifically the diastolic blood pressure recorded in millimeters of mercury (mm Hg). Further insights into body composition are offered by Skin Thickness, indicating the triceps skin fold thickness in millimeters (mm), and Insulin, which reflects the serum insulin level at 2 hours, measured in micro-units per milliliter (\u0026micro;U/ml). The widely recognized BMI a standardized measure of body fat. Moreover, the DiabetesPedigreeFunction offers a quantitative assessment of the genetic predisposition to diabetes based on the patient\u0026apos;s family history. Finally, Age represents the patient\u0026apos;s current age in years.\u003cspan dir=\"RTL\"\u003e\u0026nbsp;\u003c/span\u003eTable 2 shows the details of Dataset 3\u003cspan dir=\"RTL\"\u003e.\u003c/span\u003e The dataset itself presents a balanced distribution of outcomes, featuring 320 instances where diabetes was diagnosed positive and 200 instances where it was not negative. The feature set is diverse, comprising a total of 16 variables. This includes 14 Boolean features, likely representing binary aspects of patient history or conditions, alongside one string feature, \u0026quot;Gender,\u0026quot; and the single numerical feature, Age, providing a well-rounded perspective for predictive modeling endeavors.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2.\u003c/strong\u003e Description of Features in Frankfurt Hospital dataset.\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003eName\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003eDescription\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003eAge\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003eAge (years)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003eSex\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003eGender of the individual\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003ePolyuria\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003eExcessive urination\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003ePolydipsia\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003eFrequent thirst\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003eSudden Weight Loss\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003eRapid and unexpected weight loss\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003eWeakness\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003eGeneral body weakness\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003e\n \u003cp\u003e7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003ePolyphagia\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003eExcessive hunger and increased appetite\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003e\n \u003cp\u003e8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003eGenital thrush\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003eInfections or rashes in the genital area\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003eVisual Blurring\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003eBlurred vision\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003e\n \u003cp\u003e10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003eItching\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003ePersistent itching of the skin\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003e\n \u003cp\u003e11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003eIrritability\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003eIncreased tendency toward irritability or mood changes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003e\n \u003cp\u003e12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003eDelayed Healing\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003eSlow wound healing\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003e\n \u003cp\u003e13\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003ePartial Paresis\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003eReduced ability to move part of the body\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003e\n \u003cp\u003e14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003eMuscle stiffness\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003eMuscle tightness or rigidity\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003e\n \u003cp\u003e15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003eAlopecia\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003eSudden hair loss\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003e\n \u003cp\u003e16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003eObesity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003eExcessive body weight or obesity\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\u003cp\u003e\u003cb\u003ePre-processing\u003c/b\u003e\u003c/p\u003e\u003cp\u003ePrior to the application of sophisticated ensemble learning techniques for classification, a crucial data preprocessing phase was meticulously executed. This preparatory stage aimed to standardize the input features and render them suitable for the subsequent modeling. A key step in this process involved the application of MinMax Scaling to each feature across all datasets. This scaling technique linearly transforms the data, ensuring that all feature values are confined within the uniform range of 0 to 1. By bringing all features to a common scale, MinMax Scaling mitigates the potential for features with larger numerical ranges to unduly influence the learning algorithms. Furthermore, Dataset 3, characterized by the presence of both Boolean and string-based categorical features, necessitated an additional layer of preprocessing. To accommodate the requirements of many machine learning algorithms, these categorical values were systematically converted into numerical representations. This conversion ensures that all input features are in a quantitative format, enabling effective processing by the learning models. The Synthetic Minority Oversampling Technique (SMOTE) is recognized as a fundamental and widely used method for addressing the class imbalance problem in machine learning\u003csup\u003e\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u003c/sup\u003e. Class imbalance is a common challenge in many real-world datasets, particularly in domains such as bioinformatics, medical imaging, fraud detection, security systems, and time series forecasting. In these contexts, the number of minority class samples is significantly lower than that of the majority class, which often leads to poor performance of models in correctly identifying minority class instances, as machine learning algorithms tend to be biased towards the majority class. The core concept of SMOTE involves generating synthetic samples for the minority class through linear interpolation between existing minority class instances in the feature space. This approach aims to increase the number of minority class samples and balance the data distribution, thereby enabling machine learning models to establish more accurate decision boundaries and improve their ability to detect minority class examples. Initially, minority class samples are extracted from the training dataset. For each minority sample, its k nearest neighbors are identified based on Euclidean distance within the feature space, where the value of k is typically set to 5 but can be adjusted according to the dataset. Synthetic samples are then generated by randomly selecting one of the k nearest neighbors, computing the difference vector between the selected neighbor and the original sample, multiplying this vector by a random number δ between 0 and 1, and adding the resulting vector to the original sample SMOTE has been widely adopted due to its effectiveness in balancing class distributions and mitigating issues arising from imbalanced datasets. Compared to simpler oversampling methods such as random replication, SMOTE enhances the diversity of minority samples and reduces the risk of overfitting. Furthermore, SMOTE has been successfully integrated with deep learning models and more complex neural networks, thereby broadening its applicability across various fields including bioinformatics, medical imaging, and time series forecasting. Overall, SMOTE stands as a powerful and standard technique for addressing class imbalance, playing a crucial role in improving the quality and robustness of machine learning models. It is widely regarded as a key preprocessing step in numerous research studies and practical applications. Figure\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. shows samples of the PID dataset before and after applying SMOTE. The total number of PID dataset samples increased from 768 to 1000 after applying the SMOTE technique. Similarly, the total number of samples in the NIDDK dataset increased from 520 to 640 following the application of SMOTE. Additionally, the total number of samples in the Frankfurt Hospital Diabetes dataset increased from 2000 to 2632 after SMOTE was applied. Following the MinMax Scaling and SMOTE, the preprocessed data was partitioned into distinct training and testing subsets. This split was performed with a ratio of 80% allocated for training the models and the remaining 20% reserved for evaluating their predictive performance on unseen data, providing a robust assessment of generalization capabilities.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003cb\u003eEnsemble Learning\u003c/b\u003e\u003c/p\u003e\u003cp\u003eEnsemble Learning is a machine learning paradigm where in multiple models, often referred to as \"weak learners,\" are trained to solve the same problem and subsequently combined to yield enhanced results. The fundamental hypothesis posits that by aggregating weak models, a more accurate and robust model can be derived. Ensemble Learning encompasses three primary methodologies: Bagging, Boosting, and Stacking. A commonality among these three approaches is their reliance on the utilization of multiple base models, which are constituent machine learning models. Bagging Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. is one of the Ensemble Learning methods that utilizes only one type of base-model by performing parallel and independent learning on each base-model, and subsequently combining them to obtain the optimal result. The models to be used in bagging ensemble learning are: Bagging, Random Forest, Extra Trees. The base-model used is Decision Tree, as it has consistently yielded the best results in previous research\u003csup\u003e\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u003c/sup\u003e. Boosting Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. is a method within Ensemble Learning that employs a single type of base model, trained in a sequential and adaptive manner. In this approach, the output of each base model depends on the performance of the previous models, and the results are subsequently combined to achieve optimal performance. The models utilized in boosting ensemble learning include: Adaptive Boosting (AdaBoost), Gradient Boosting Machines (GBM), Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), CatBoost The base model used in this study is the Decision Tree, as it has consistently demonstrated superior performance in previous research\u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eStacking Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e. is an Ensemble Learning method that employs multiple base models, each trained in parallel and independently\u003csup\u003e\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e\u003c/sup\u003e. The outputs of these base models are then combined using a meta-learning algorithm to produce a final predictive result from the combination of the base models. he architecture of the stacking model involves two or more base models, commonly referred to as level-0 models, and a meta-model, known as the level-1 model, which combines the predictions generated by the base models. In this stacking ensemble learning experiment, five base models are utilized: Logistic Regression, Support Vector Machine (SVC), Naive Bayes (Gaussian NB), K-Nearest Neighbors (KNN), Decision Tree The meta-model employed in this study is Logistic Regression. The proposed framework is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003cb\u003eEvaluation metrics\u003c/b\u003e\u003c/p\u003e\u003cp\u003eThe number of diseased individuals correctly identified by the system is referred to as True Positive (TP). The number of healthy individuals correctly identified by the system is considered True Negative (TN). The number of healthy individuals incorrectly classified as diseased is referred to as False Positive (FP), and the number of diseased individuals incorrectly classified as healthy is considered False Negative (FN).\u003c/p\u003e\u003cp\u003eThe accuracy metric can be evaluated after the model has been trained. In this chapter, we will analyze the results of the proposed model. Accuracy is the primary metric used to assess the model's performance in predicting true positive and true negative cases. Accuracy can be calculated using Eq.\u0026nbsp;(\u003cspan refid=\"Equ1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003cdiv id=\"Equ1\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ1\" name=\"EquationSource\"\u003e\n$$\\:ACC=\\left(\\frac{TP+TN}{\\left(TP+TN+FP+FN\\right)}\\right)$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e1\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eThe ratio of true positive observations to the total number of predicted positive cases is referred to as precision. A precision value of 1 indicates that the proposed model performs well. Precision can be calculated using Eq.\u0026nbsp;(\u003cspan refid=\"Equ2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003cdiv id=\"Equ2\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ2\" name=\"EquationSource\"\u003e\n$$\\:PR=\\left(\\frac{TP}{TP+FP}\\right)$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e2\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eThe recall metric can be defined as sensitivity, which indicates the classifier\u0026rsquo;s ability to identify all positive instances. Recall can be calculated using Eq.\u0026nbsp;(\u003cspan refid=\"Equ3\" class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003cdiv id=\"Equ3\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ3\" name=\"EquationSource\"\u003e\n$$\\:REC=\\left(\\frac{TP}{TP+FN}\\right)$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e3\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eIn this study, three datasets each initially subjected to preprocessing, subsequently balanced using the SMOTE, and partitioned into training and testing subsets were employed for the classification task, which was conducted through the application of three ensemble learning approaches and their respective methodologies.\u003c/p\u003e\u003cp\u003eThe performance of the PID dataset was systematically evaluated under two experimental conditions: prior to the application of the SMOTE technique and subsequent to its implementation. Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e presents the evaluation metrics of the models before applying the SMOTE technique, while Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e presents the evaluation metrics of the models after applying the SMOTE technique.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003ethe evaluation metrics of the models on the PID dataset before SMOTE.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"6\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eEnsemble Method\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eModel\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eAccuracy\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003ePrecision\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003eRecall\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003e1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003eBagging\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eBagging\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e77.92%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e73.96%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e73.65%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eRandom Forest\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e79.87%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e76.36%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e75.37%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eExtra Trees\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e79.87%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e76.28%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e75.96%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"4\" rowspan=\"5\"\u003e\u003cp\u003e2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\" morerows=\"4\" rowspan=\"5\"\u003e\u003cp\u003eBoosting\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eAdaptive Boosting\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e78.57%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e74.85%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e73.92%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eGradient Boosting\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e80.51%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e77.03%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e77.03%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eExtreme Gradient Boosting\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e80.51%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e77.07%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e78.82%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eLight Gradient Boosting\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e79.87%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e76.26%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e77.16%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eCat Boosting\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e82.46%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e79.54%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e78.43%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eStacking\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eStacked Generalization\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e80.51%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e77.63%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e74.64%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003ethe evaluation metrics of the models on the PID dataset after SMOTE.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"6\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eEnsemble Method\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eModel\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eAccuracy\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003ePrecision\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003eRecall\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003e1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003eBagging\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eBagging\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e81.50%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e81.65%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e81.36%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eRandom Forest\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e85.00%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e84.97%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e84.06%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eExtra Trees\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e87.50%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e87.49%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e87.47%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"4\" rowspan=\"5\"\u003e\u003cp\u003e2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\" morerows=\"4\" rowspan=\"5\"\u003e\u003cp\u003eBoosting\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eAdaptive Boosting\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e81.00%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e80.97%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e81.05%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eGradient Boosting\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e83.00%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e82.95%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e82.95%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eExtreme Gradient Boosting\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e81.50%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e81.76%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e81.72%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eLight Gradient Boosting\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e83.50%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e83.56%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e83.63%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eCat Boosting\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e83.50%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e83.47%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e83.43%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eStacking\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eStacked Generalization\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e87.00%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e86.69%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e86.96%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eBefore SMOTE the highest accuracy was achieved by CatBoost 82.46%, outperforming other boosting and bagging models. Overall, the performance of most models ranged between 77% and 80%, with precision and recall values consistently within a similar range (approximately 73%\u0026ndash;79%). After SMOTE all models exhibited a noticeable increase in performance across accuracy, precision, and recall metrics. The most significant improvements were observed for Extra Trees 87.50% and Stacking 87.00%, which achieved the highest overall performance. Even models with relatively lower performance prior to SMOTE, such as Adaptive Boosting, demonstrated marked improvement, reaching accuracy levels above 81%.\u003c/p\u003e\u003cp\u003eThe application of the SMOTE technique significantly enhanced the balance of the dataset, resulting in overall improvements in classification performance. In particular, Extra Trees and Stacking emerged as the most effective ensemble methods on the balanced dataset, highlighting the superior capability of randomization-based methods and model-combination strategies in handling imbalanced data. The Frankfurt Hospital in Germany, contains the same features as PID but consists of a larger number of records. The experimental results of ensemble learning on Frankfurt Hospital dataset are in Table\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003ethe evaluation metrics of the models on the Frankfurt Hospital dataset after SMOTE. \u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"6\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eEnsemble Method\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eModel\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eAccuracy\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003ePrecision\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003eRecall\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003e1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003eBagging\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eBagging\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e97.25%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e98.05%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e95.70%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eRandom Forest\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e99.50%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e99.63%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e99.21%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eExtra Trees\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e98.50%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e98.47%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e98.06%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"4\" rowspan=\"5\"\u003e\u003cp\u003e2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\" morerows=\"4\" rowspan=\"5\"\u003e\u003cp\u003eBoosting\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eAdaptive Boosting\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e76.75%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e73.33%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e72.38%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eGradient Boosting\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e86.00%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e85.55%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e81.43%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eExtreme Gradient Boosting\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e99.25%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e99.23%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e99.13%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eLight Gradient Boosting\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e99.25%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e99.23%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e99.13%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eCat Boosting\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e95.75%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e96.24%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e93.79%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eStacking\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eStacked Generalization\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e98.50%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e98.09%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e98.28%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\n\u003cp\u003eNIDDK dataset obtained from a hospital in Sylhet, Bangladesh, contains features that differ from those of other Datasets. The ensemble learning experiments conducted on NIDDK dataset are in Table 6.\u003c/p\u003e\n\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab6\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 6\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003ethe evaluation metrics of the models on NIDDK dataset after SMOTE.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"6\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eEnsemble Method\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eModel\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eAccuracy\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003ePrecision\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003eRecall\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003e1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003eBagging\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eBagging\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e98.07%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e97.61%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e98.43%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eRandom Forest\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e99.03%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e98.78%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e99.21%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eExtra Trees\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e100%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e100%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e100%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"4\" rowspan=\"5\"\u003e\u003cp\u003e2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\" morerows=\"4\" rowspan=\"5\"\u003e\u003cp\u003eBoosting\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eAdaptive Boosting\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e95.19%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e95.66%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e94.21%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eGradient Boosting\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e99.03%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e98.78%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e99.21%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eExtreme Gradient Boosting\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e99.03%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e98.78%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e99.21%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eLight Gradient Boosting\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e100%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e100%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e100%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eCat Boosting\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e100%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e100%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e100%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eStacking\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eStacked Generalization\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e99.03%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e98.87%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e99.21%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eThe experimental results of ensemble learning methods on the Frankfurt Hospital and NIDDK datasets are summarized in Tables\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e and \u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e6\u003c/span\u003e, respectively. A comparative analysis of the two datasets reveals notable differences in the performance of ensemble algorithms. On the Frankfurt Hospital dataset, the performance of ensemble models varied significantly across different methods. Bagging-based techniques demonstrated competitive performance, with Random Forest achieving the highest accuracy of 99.50%, followed closely by Extra Trees 98.50%. In contrast, some boosting algorithms, particularly Adaptive Boosting 76.75% and Gradient Boosting 86.00%, exhibited relatively low performance. Nevertheless, advanced boosting methods such as Extreme Gradient Boosting 99.25% and Light Gradient Boosting 99.25% performed remarkably well, alongside CatBoost 95.75%. Stacking also achieved a strong overall accuracy of 98.50%. These findings indicate that while certain boosting approaches may be sensitive to the data distribution, tree-based ensembles such as Random Forest, XGBoost, and LightGBM maintain robust generalization performance. On the other hand, the NIDDK dataset demonstrated consistently higher results across all ensemble methods. Bagging, Random Forest, and Extra Trees all achieved outstanding performance, with Extra Trees reaching a perfect score of 100% in all evaluation metrics. Similarly, boosting methods such as Gradient Boosting, XGBoost, and LightGBM reached accuracies above 99%, with both LightGBM and CatBoost attaining perfect classification performance. Even Adaptive Boosting, which showed poor performance on the Frankfurt dataset, improved substantially, achieving 95.19% accuracy. Stacking also achieved a high accuracy of 99.03%. while the Frankfurt dataset exhibited larger variability among ensemble models, the NIDDK dataset yielded consistently superior and nearly perfect results across all methods. This suggests that the NIDDK dataset may contain more distinctive patterns or separable features that are effectively leveraged by ensemble algorithms, whereas the Frankfurt dataset presents a more challenging classification task, particularly for certain boosting techniques.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThe findings of this study demonstrate that the integration of ensemble learning methods with the SMOTE technique can significantly enhance the classification performance of diabetes patients, particularly in imbalanced datasets. Analysis across three datasets\u0026mdash;Pima Indians Diabetes, Frankfurt Hospital Diabetes, and Sylhet Hospital Diabetes (NIDDK)\u0026mdash;revealed notable differences in data characteristics and the behavior of machine learning algorithms. Overall, Boosting-based algorithms outperformed Bagging and Stacking approaches. This advantage was particularly evident in the Pima dataset, which was smaller and highly affected by class imbalance. Following SMOTE application, model performance improved substantially, with Gradient Boosting, Extreme Gradient Boosting, and CatBoost achieving the highest accuracy of 81.82%. These results underscore the critical importance of data preprocessing and addressing class imbalance to improve model generalizability. In the Frankfurt Hospital dataset, while Bagging methods\u0026mdash;especially Random Forest (99.50%) and Extra Trees (98.50%)\u0026mdash;demonstrated strong performance, certain Boosting methods, such as Adaptive Boosting and Gradient Boosting, showed comparatively lower accuracy. Conversely, more advanced tree-based Boosting algorithms, including LightGBM and XGBoost, achieved accuracies above 99%, highlighting their ability to capture complex patterns in clinical data effectively. For the Sylhet Hospital (NIDDK) dataset, nearly all ensemble learning methods achieved near-perfect performance. Algorithms such as Extra Trees, LightGBM, and CatBoost attained 100% accuracy, precision, and recall, effectively distinguishing between healthy and diabetic cases. Even Adaptive Boosting, which performed poorly on the Frankfurt dataset, improved substantially to reach 95.19% accuracy in this dataset. These findings indicate that variations in data quality, feature distribution, and class separability play a pivotal role in the success or failure of ensemble algorithms. Comparative analysis across the three datasets suggests that dataset size, quality, and feature distribution directly influence model performance. While the Frankfurt dataset posed a more challenging classification task for certain Boosting methods, the NIDDK dataset exhibited more distinctive patterns, allowing advanced algorithms to achieve nearly perfect classification. This emphasizes the importance of selecting appropriate training datasets in combination with powerful algorithms for optimal performance. From a practical perspective, these results have significant implications for precision medicine and clinical decision-making. Ensemble-based models not only facilitate early detection of diabetes but can also serve as decision-support tools for risk assessment, resource allocation, and the design of individualized interventions. Nevertheless, several limitations should be acknowledged. First, while SMOTE effectively balanced the datasets, synthetic sample generation may occasionally lead to overfitting. Second, the results are dependent on the quality of the datasets used and may vary when applied to larger or more heterogeneous populations. Finally, this study considered only three datasets, and full generalizability requires validation on multi-center and larger-scale datasets. Future research should explore more advanced hybrid approaches, including the integration of deep learning with ensemble methods. Moreover, attention to model interpretability is crucial in the medical domain to ensure clinicians\u0026rsquo; confidence in deploying these tools in real-world clinical settings.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAuthor contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eS.A.J.: Conceptualization, Methodology, Investigation, Implementation, Visualization, Analysis and interpretation, Data curation, Writing - original draft. A.K.: Methodology, Investigation, Writing - review \u0026amp; editing, Supervision, Data curation, Validation, Project administration, Formal analysis. All authors took part in the work described in this paper. All authors read and approved the final manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis research did not receive any specific grant from funding agencies in the public, commercial, or not-forprofit sectors.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics approval\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis manuscript, or a large part of it, has not been published, was not, and is not being submitted to any other journal. All text and graphics, except for those marked with sources, are original works of the authors. All authors each made a significant contribution to the research reported and have read and approved the submitted manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAdditional information\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eCorrespondence and requests for materials should be addressed to sajjad Aghasi Javid at [email protected]\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eReprints and permissions information\u003c/strong\u003e is available at www.nature.com/reprints.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePublisher\u0026rsquo;s note\u003c/strong\u003e Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.\u0026nbsp;\u003c/p\u003e\u003cp\u003e\u003cstrong\u003eData Availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eData is available from the first author ([email protected]) upon a reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eHameed, I. et al. Type 2 diabetes mellitus: from a metabolic disorder to an inflammatory condition. \u003cem\u003eWorld J. diabetes\u003c/em\u003e. \u003cb\u003e6\u003c/b\u003e, 598 (2015).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eUmpierrez, G. E. \u003cem\u003eCardiovascular Outcomes of Treatments available for Patients with Type 1 and 2 Diabetes, An Issue of Endocrinology and Metabolism Clinics of North America, E-Book: Cardiovascular Outcomes of Treatments available for Patients with Type 1 and 2 Diabetes, An Issue of Endocrinology and Metabolism Clinics of North America, E-Book\u003c/em\u003e. Vol. 47Elsevier Health Sciences, (2018).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRoglic, G. \u003cem\u003eGlobal report on diabetes\u003c/em\u003e (World Health Organization, 2016).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhang, H. \u0026amp; Shi, H. Construction of a prediction model for coronary heart disease in type 2 diabetes mellitus: a cross-sectional study. \u003cem\u003eSci. Rep.\u003c/em\u003e \u003cb\u003e15\u003c/b\u003e, 7003 (2025).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAbousaber, I., Abdallah, H. F. \u0026amp; El-Ghaish, H. Robust predictive framework for diabetes classification using optimized machine learning on imbalanced datasets. \u003cem\u003eFront. Artif. Intell.\u003c/em\u003e \u003cb\u003e7\u003c/b\u003e, 1499530 (2025).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHassan, M. M., Mollick, S. \u0026amp; Yasmin, F. An unsupervised cluster-based feature grouping model for early diabetes detection. \u003cem\u003eHealthc. Analytics\u003c/em\u003e. \u003cb\u003e2\u003c/b\u003e, 100112 (2022).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSampath, P. et al. Robust diabetic prediction using ensemble machine learning models with synthetic minority over-sampling technique. \u003cem\u003eSci. Rep.\u003c/em\u003e \u003cb\u003e14\u003c/b\u003e, 28984 (2024).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSmith, J. W., Everhart, J. E., Dickson, W. C., Knowler, W. C. \u0026amp; Johannes, R. S. in \u003cem\u003eProceedings of the annual symposium on computer application in medical care.\u003c/em\u003e 261.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBhoi, S. K. et al. Prediction of diabetes in females of pima Indian heritage: a complete supervised learning approach. \u003cem\u003eTurkish J. Comput. Math. Educ.\u003c/em\u003e \u003cb\u003e12\u003c/b\u003e, 3074\u0026ndash;3084 (2021).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAgatsa, D. A., Rismala, R. \u0026amp; Wisesty, U. N. Klasifikasi Pasien Pengidap Diabetes menggunakan Metode Support Vector Machine. \u003cem\u003eeProceedings of Engineering\u003c/em\u003e 7 (2020).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKaratsiolis, S. \u0026amp; Schizas, C. N. in \u003cem\u003eIEEE 12th international conference on bioinformatics \u0026amp; bioengineering (BIBE).\u003c/em\u003e 139\u0026ndash;144 (IEEE). 139\u0026ndash;144 (IEEE). (2012).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMaulidina, F. et al. in \u003cem\u003eJournal of Physics: Conference Series.\u003c/em\u003e 012006 (IOP Publishing).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDaanouni, O., Cherradi, B. \u0026amp; Tmiri, A. in \u003cem\u003eProceedings of the 4th International Conference on Smart City Applications.\u003c/em\u003e 1\u0026ndash;6.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNai-Arun, N. \u0026amp; Sittidech, P. Ensemble learning model for diabetes classification. \u003cem\u003eAdv. Mater. Res.\u003c/em\u003e \u003cb\u003e931\u003c/b\u003e, 1427\u0026ndash;1431 (2014).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePradhan, G., Pradhan, R. \u0026amp; Khandelwal, B. in \u003cem\u003eSoft Computing Techniques and Applications: Proceeding of the International Conference on Computing and Communication (IC3\u003c/em\u003e). 553\u0026ndash;561 (Springer).). 553\u0026ndash;561 (Springer). (2020).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKumari, S., Kumar, D. \u0026amp; Mittal, M. An ensemble approach for classification and prediction of diabetes mellitus using soft voting classifier. \u003cem\u003eInt. J. Cogn. Comput. Eng.\u003c/em\u003e \u003cb\u003e2\u003c/b\u003e, 40\u0026ndash;46 (2021).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSarwar, M. A., Kamal, N., Hamid, W. \u0026amp; Shah, M. A. in \u003cem\u003e24th international conference on automation and computing (ICAC).\u003c/em\u003e 1\u0026ndash;6 (IEEE). 1\u0026ndash;6 (IEEE). (2018).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDey, S. K., Hossain, A. \u0026amp; Rahman, M. M. in \u003cem\u003e2018 21st international conference of computer and information technology (ICCIT).\u003c/em\u003e 1\u0026ndash;5 (IEEE).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMir, A. \u0026amp; Dhage, S. N. in \u003cem\u003efourth international conference on computing communication control and automation (ICCUBEA).\u003c/em\u003e 1\u0026ndash;6 (IEEE). 1\u0026ndash;6 (IEEE). (2018).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSaru, S. \u0026amp; Subashree, S. Analysis and prediction of diabetes using machine learning. \u003cem\u003eInternational J. Emerg. Technol. innovative engineering\u003c/em\u003e \u003cb\u003e5\u003c/b\u003e (2019).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSonar, P. \u0026amp; JayaMalini, K. in \u003cem\u003e3rd International Conference on Computing Methodologies and Communication (ICCMC).\u003c/em\u003e 367\u0026ndash;371 (IEEE). 367\u0026ndash;371 (IEEE). (2019).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWei, S., Zhao, X. \u0026amp; Miao, C. in \u003cem\u003e2018 IEEE 4th World Forum on Internet of Things (WF-IoT).\u003c/em\u003e 291\u0026ndash;295 (IEEE).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFaruque, M. F. \u0026amp; Sarker, I. H. in \u003cem\u003e2019 international conference on electrical, computer and communication engineering (ECCE).\u003c/em\u003e 1\u0026ndash;4 (IEEE).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJain, B., Ranawat, N., Chittora, P., Chakrabarti, P. \u0026amp; Poddar, S. (Elsevier, (2021).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMung, P. S. \u0026amp; Phyu, S. \u003cem\u003eEnsemble learning method for enhancing healthcare classification\u003c/em\u003e (MERAL Portal, 2020).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e\u003cem\u003eUCI Machine Learning. Pima Indians Diabetes Database\u003c/em\u003e, \u0026lt; (2016). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.kaggle.com/uciml/pima-indians-diabetes-database\u003c/span\u003e\u003cspan address=\"https://www.kaggle.com/uciml/pima-indians-diabetes-database\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJohn. \u003cem\u003ediabetes\u003c/em\u003e, (2017). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.kaggle.com/johndasilva/diabetes\u003c/span\u003e\u003cspan address=\"https://www.kaggle.com/johndasilva/diabetes\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDutta, I. \u003cem\u003eEarly Stage Diabetes Risk Prediction Dataset\u003c/em\u003e, \u0026lt; (2020). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.kaggle.com/ishandutta/early-stage-diabetes-risk-prediction-dataset\u003c/span\u003e\u003cspan address=\"https://www.kaggle.com/ishandutta/early-stage-diabetes-risk-prediction-dataset\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMansourifar, H. \u0026amp; Shi, W. Deep synthetic minority over-sampling technique. \u003cem\u003earXiv preprint arXiv\u003c/em\u003e:.\u003cem\u003e09788\u003c/em\u003e (2020). (2020). (2003).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFerrouhi, E. M. \u0026amp; Bouabdallaoui, I. A comparative study of ensemble learning algorithms for high-frequency trading. \u003cem\u003eSci. Afr.\u003c/em\u003e \u003cb\u003e24\u003c/b\u003e, e02161 (2024).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eOliullah, K. et al. A stacked ensemble machine learning approach for the prediction of diabetes. \u003cem\u003eJ. Diabetes Metabolic Disorders\u003c/em\u003e. \u003cb\u003e23\u003c/b\u003e, 603\u0026ndash;617 (2024).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLaila, U. E., Mahboob, K., Khan, A. W., Khan, F. \u0026amp; Taekeun, W. An ensemble approach to predict early-stage diabetes risk using machine learning: An empirical study. \u003cem\u003eSensors\u003c/em\u003e \u003cb\u003e22\u003c/b\u003e, 5247 (2022).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCendani, L. M. \u0026amp; Wibowo, A. Perbandingan Metode ensemble learning pada klasifikasi penyakit diabetes. \u003cem\u003eJurnal Masyarakat Informatika\u003c/em\u003e. \u003cb\u003e13\u003c/b\u003e, 33\u0026ndash;44 (2022).\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Bagging, Boosting, Stacking, Diabetes, Ensemble Learning","lastPublishedDoi":"10.21203/rs.3.rs-7775507/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7775507/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"Diabetes mellitus is a prevalent metabolic disorder characterized by chronic hyperglycemia and associated with severe complications. Accurate early detection is essential for effective management and prevention of disease progression. This study systematically evaluates the performance of three ensemble learning approaches Bagging, Boosting, and Stacking on three benchmark diabetes datasets: Pima Indians Diabetes, Frankfurt Hospital Diabetes, and Sylhet Hospital Diabetes (NIDDK). Class imbalance, a common challenge in these datasets, was addressed using the Synthetic Minority Oversampling Technique (SMOTE) during preprocessing to enhance model stability and classification reliability. Experimental results indicate that Boosting-based methods consistently outperform Bagging and Stacking. On the Pima dataset, Gradient Boosting, Extreme Gradient Boosting, and CatBoost achieved a maximum accuracy of 81.82%. On the Frankfurt dataset, Light Gradient Boosting reached 99.25% accuracy, while on the NIDDK dataset, Light Gradient Boosting and CatBoost attained perfect accuracy (100%). These findings highlight the effectiveness of integrating SMOTE with Boosting-based ensemble models to mitigate class imbalance and improve diabetes classification. The results underscore the importance of both data preprocessing and algorithm selection in achieving high predictive performance, with significant implications for precision medicine and clinical decision support.","manuscriptTitle":"Evaluating the Performance of Ensemble Learning Methods in Diabetes Disease Classification","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-11-03 14:40:03","doi":"10.21203/rs.3.rs-7775507/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-11-25T18:10:48+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-11-21T13:04:36+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"171257081199747048912503422436744346647","date":"2025-11-14T20:19:42+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-11-14T18:21:31+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"118285148597829266399557171833386547362","date":"2025-10-23T17:41:34+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-10-23T08:16:50+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-10-15T19:22:26+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-10-13T07:06:54+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-10-13T07:06:36+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2025-10-03T17:46:36+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"005581b5-de55-4951-b218-d4c798022a9a","owner":[],"postedDate":"November 3rd, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"in-revision","subjectAreas":[{"id":57286909,"name":"Biological sciences/Computational biology and bioinformatics"},{"id":57286910,"name":"Health sciences/Diseases"},{"id":57286911,"name":"Health sciences/Endocrinology"},{"id":57286912,"name":"Health sciences/Health care"},{"id":57286913,"name":"Physical sciences/Mathematics and computing"},{"id":57286914,"name":"Health sciences/Medical research"}],"tags":[],"updatedAt":"2026-05-18T16:39:23+00:00","versionOfRecord":[],"versionCreatedAt":"2025-11-03 14:40:03","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7775507","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7775507","identity":"rs-7775507","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}