An Evaluation of Machine Learning Categories for Diabetes Prediction and Detection in Libya: A Comparative Study

An Evaluation of Machine Learning Categories for Diabetes Prediction and Detection in Libya: A Comparative Study | 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 Research Article An Evaluation of Machine Learning Categories for Diabetes Prediction and Detection in Libya: A Comparative Study Sama Tarek Jadidi, Abdulbaset Mustafa Goweder, Mohamed Hadi Mohamed Abdelhamid This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7687597/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background: Diabetes Mellitus is a growing global health concern, projected to affect over 1.31 billion people by 2050. Early detection is vital, and machine learning offers a promising tool for predicting and managing the disease. Aim: This study aimed to introduce a structured classification of ML algorithms into three categories and to evaluate their performance in predicting diabetes using locally collected patient data. Methods: A dataset of 806 participants (403 diabetic and 403 non-diabetic) was analyzed using attributes such as sex, age, body mass index, blood glucose, blood pressure, diabetes pedigree function, and number of pregnancies (females only). ML algorithms were grouped into three categories: Simple Computational (Logistic Regression, Naïve Bayes), Tree-based (Random Forests, Gradient Boosted Trees), and Margin-based (Support Vector Machines, Fast Large Margin). Data were partitioned into training, validation, and testing sets using stratified sampling and cross-validation. Performance was assessed using accuracy, error rate, precision, recall, specificity, and F-measure. Results: Tree-based algorithms outperformed other categories, with Gradient Boosted Trees achieving the highest accuracy (97.8%), followed by Random Forests (97.5%). This category also achieved superior specificity, precision, and F-measure. In contrast, Simple Computational algorithms showed the highest sensitivity (Logistic Regression 99.3%, Naïve Bayes 98.8%), effectively identifying true positive cases. Conclusion: The study’s classification framework provides a systematic basis for comparing ML models, highlighting the strengths of each category. It offers a foundation for hybrid approaches that combine high accuracy with strong sensitivity, supporting enhanced diagnostic accuracy and improved clinical decision-making. Diabetes Mellitus Machine Learning Categories Healthcare Clinical Prediction models Predictive analytics Figures Figure 1 Figure 2 Figure 3 Introduction Diabetes mellitus is a chronic disease that is witnessing an increasing prevalence worldwide. This disease leads to serious complications, including cardiovascular diseases, kidney failure, retinopathy, neuropathy, and other complications that may result in limb amputation. Effective management and early prediction of diabetes are essential to reduce these risks and improve the quality of healthcare (Holt et al., 2024 ; Ong et al., 2023; WHO, 2023 ). On the other hand, artificial Intelligence (AI) is a multidisciplinary branch of computer science that focuses on developing systems capable of simulating human intelligence and performing tasks requiring advanced cognitive abilities (Pragyna Karmakar et al., 2024; Sheikh et al., 2023 ). Significantly, AI has made substantial contributions to healthcare by enhancing clinical decision-making accuracy. Through its ability to analyze extensive datasets and identify complex patterns and relationships with high precision, AI empowers medical professionals to make more accurate and effective treatment decisions (Davenport & Kalakota, 2019 ; Khosravi et al., 2024 ). Moreover, in the healthcare domain, AI showcases immense potential in improving the diagnosis, prediction, and management of chronic diseases, including conditions such as diabetes, hypertension, heart disease, and kidney disease. By integrating AI into medical practices, healthcare providers can leverage its capabilities to optimize patient care and enhance overall health outcomes (Islam et al., 2024 ; Khalifa & Albadawy, 2024 ; Liu et al., 2025 ). In fact, Machine learning (ML), a key area of artificial intelligence (AI), focuses on analyzing, classifying, and predicting outcomes based on patterns in complex datasets. In healthcare—especially in diabetes management—ML has emerged a transformative tool for early diagnosis, predicting disease progression, and optimizing treatment plans(Corrao et al., 2025 ; Kanagarathinam et al., 2024 ; Wu, 2024 ) by using ML techniques, various medical parameters such as blood glucose levels, blood pressure, body mass index (BMI), and genetic traits can be systematically examined to identify individuals at risk of developing diabetes. One of the most significant uses of ML in this field is its ability to detect early signs of the disease, enabling timely intervention and reducing the risk of complications (Arkoudis & Papadakos, 2025). Additionally, ML algorithms are used to predict diabetes-related complications, including diabetic retinopathy and cardiovascular issues, and to support personalized treatment plans by customizing recommendations for each patient, ultimately improving treatment results and reducing associated risks (Arkoudis & Papadakos, 2025; Oikonomou & Khera, 2023 ). Several studies have shown that the prediction and diagnosis of diabetes, using techniques such as machine learning, deep learning, and neural networks, range from 69% to 98.38% in accuracy (Table .1). Lai et al. ( 2019 ) implemented a study to develop a predictive model for identifying individuals at risk of developing diabetes (Lai et al., 2019 ). The study applied Logistic Regression and Gradient Boosting Machine models and compared their performance with Decision Tree and Random Forest models. In the same vein, the analysis relied on data from the Canadian CPCSSN database and the PIMA Indians dataset (PIDD). The results demonstrated that the GBM model outperformed other models with an AROC of 84.7% and a sensitivity of 71.6%, while LR achieved an AROC of 84.0% and a sensitivity of 73.4%. Furthermore, Ullah et al., ( 2022 ) conducted a study to detect diabetes using machine learning techniques (Ullah et al., 2022 ). The applied algorithms included K-Nearest Neighbors (KNN), Random Forest, XGBoost, Bagging, and AdaBoost, alongside the SMOTE-ENN technique to address data imbalance. In fact, the study used data from the publicly available Behavioral Risk Factor Surveillance System (BRFSS) database. Importantly, the findings demonstrated that the KNN algorithm achieved the highest accuracy at 98.38%. Likewise, Khaleel & Al-bakry ( 2023 ) showed that to develop a model capable of predicting diabetes diagnosis Click or tap here to enter text.(Khaleel & Al-Bakry, 2023 ). Several machine learning algorithms were applied, including Logistic Regression, Naive Bayes, and K-Nearest Neighbors, using the Pima Indian Diabetes Database (PIDD). Their comparative results revealed that LR achieved the highest accuracy at 94%, followed by NB at 79%, and KNN at 69%. Similarly, Alzboon et al., (2023) evaluated the effectiveness of several machine learning algorithms for the early prediction of diabetes (Alzboon et al., 2023). The algorithms included Logistic Regression, Decision Tree, Random Forest, K-Nearest Neighbors, Naive Bayes, Support Vector Machine, Gradient Boosting, and a neural network. The analysis was conducted using the Pima Indian Diabetes Database. According to their findings, the neural network algorithm outperformed others, achieving an accuracy of 78.57%, followed by Random Forest at 76.30%, and Support Vector Machine at 73.9%. Moreover, Wang et al. ( 2024 ) carried out a study focusing on predicting diabetes risk using a LASSO-regularized regression model, based on data from the Alibaba Cloud Tianchi competition (Wang et al., 2024 ). The results indicated that the LASSO model achieved an AUC of 84.8%, employing a ten-fold cross-validation approach. In a recent study, Khaledi et al. (2025) conducted a study aimed at developing machine learning models to predict diabetes. In this context, the study utilized Logistic Regression, Decision Tree, Random Forest, bagging methods, boosting techniques (including AdaBoost and Gradient Boosted Decision Trees (GBDT)), and Support Vector Machine (SVM) (Khaledi et al., 2025). The data for model development were obtained from the Shahedieh cohort. Notably, the findings revealed that the AdaBoost model achieved the highest accuracy, reaching 86.2%. In a similar vein, Ghazizadeh (2025) evaluated the performance of various machine learning models for diabetes prediction using the PIMA Indians dataset Ghazizadeh 2025). The models included Random Forest, Support Vector Machine, Logistic Regression, Decision Tree, Naive Bayes, and K-Nearest Neighbors. According to their comparative analysis, the Random Forest model achieved the highest accuracy at 84%, while both Support Vector Machines and Logistic Regression models recorded an accuracy of 82%. Furthermore, Kiran et al. ( 2025 ) conducted a comprehensive systematic review of literature spanning 33 years on diabetes prediction using machine learning techniques (Kiran et al., 2025 ). The review indicated that Decision Tree, Random Forest, Gradient Boosting Machines (GBMs), and Neural Networks (NNs) consistently demonstrated strong predictive performance. Furthermore, it highlighted that Tree-based ensemble methods, particularly Random Forest and Gradient Boosting, often outperformed other algorithms in terms of prediction accuracy. In this context, this study aims to detect and predict DM using Simple Computational, Tree-based, and Margin-based algorithms. It compares the performance of the built models to determine which category of algorithms is most effective for the task of DM detection and prediction. Materials and Methods In this research study, six different machine learning algorithms were adopted and applied to build different models for the detection and prediction of diabetes among Libyan patients. These machine learning algorithms belong to three main categories: Simple Computational algorithms, Tree-based algorithms, and Margin-based algorithms. The built models are evaluated to assess their performance in order to compare the three main categories. Study Population In this study, the dataset was systematically collected, comprising a total of 806 patients classified into two groups: 403 non-diabetic individuals (control group) and 403 diabetic individuals (case group). This balanced distribution ensures better performance in predicting the probability of diabetes. To improve the accuracy of the analysis and ensure more reliable predictions, the selection of features for this study was based on previous literature, supplemented by consultations with endocrinologists, diabetes specialists, and physicians from other specialties. Subsequently, a set of features deemed essential and highly impactful for diabetes prediction was selected. The features included in this study comprised age (restricted to adults), sex, and the number of pregnancies for female participants. Additionally, the study incorporated the measurement of Body Mass Index (BMI), which was calculated based on height and weight data. Clinical indicators such as fasting blood glucose level and diastolic blood pressure were also included, given their well-established significance in assessing the risk of developing diabetes. Furthermore, the Pedigree Diabetes Function (PDF) was incorporated, which accounts for genetic factors by determining whether a participant has a parent or sibling diagnosed with diabetes. Finally, the dataset included the diabetes status of each participant, which served as the target variable for the prediction process. In accordance with the features adopted in this study, the distribution of sex features among participants was determined, with 359 males and 447 females. This distribution reflects a roughly balanced representation between the two groups, thereby attempting to enhance the accuracy of the study's results. All records related to individuals under the age of 18 were excluded to ensure consideration only of adults. Additionally, any data obtained from patient files were removed to ensure the accuracy and reliability of the dataset (Fig. 1 ). Data Collection and Sources The data were gathered by a team of 12 healthcare professionals (Tripoli Teaching Hospital, National Center for Diabetes and Endocrinology, National Center for Disease Control - Souk Jomaa Branch, Janzour Kidney Dialysis Services Center, Qasr Alhamra Pharmacy, and Al-Qasr Pharmacy). professionals, including Physicians, Nurses, Pharmacists, and Medical Students. Data collection started in February 2023 and proceeded until August 2024. The goal was to gather the necessary number of data points while ensuring their accuracy and reliability. Additionally, efforts were made to balance the sample distribution to accurately represent the targeted groups. Moreover, the Kobo Toolbox software was used for data collection ( https://ee.kobotoolbox.org/x/nIj5eTMb ), as it is an effective tool for gathering medical field data. It allows data collection using mobile devices, which significantly facilitates data management. After the data was collected, it was stored within the Kobo software. Subsequently, the data was transferred to Microsoft Excel, where basic calculations were applied to ensure accurate analysis and further processing. Study Questionnaire The questionnaire began with a screening question to identify participants with a prior diabetes diagnosis. It then consisted of three thematic sections: the first collected demographic and personal information (such as age, sex, marital status, number of children, nationality, and occupation) to provide context for risk factors; the second gathered key health indicators (height, weight, fasting blood glucose, diastolic blood pressure) to explore biological causes; and the third examined family history by noting diabetes in close relatives and the number of affected siblings to evaluate hereditary and genetic influences. Data Cleaning and Calculation Methods Data cleaning is considered one of the essential steps to ensure the accuracy of the results. Missing data were addressed by deleting records containing any incomplete information. Similarly, outliers or errors were handled in the same manner by removing the entire record. Furthermore, data formatting was verified to ensure consistency across different measurement units. In addition to the cleaning process, computational methods were applied to calculate the BMI and PDF indicators: The BMI was calculated using the standard formula: \(\:BMI=\frac{Weight\left(kg\right)}{{height\left(m\right)}^{2}}\) Eq. (1 ) The Pedigree Diabetes Function (PDF) is a risk factor derived from family history and was calculated using the following formula: \(\:PDF=\left(0.3*\:Siblings\right)\:+\:\left(0.4\:*\:Mother\right)\:+\:\left(0.3\:*\:Father\right)\:\) Eq. (2 ) Table 1 shows the values to be substituted for the variables (Siblings, Mother, and Father) of Eq. (2). The collected dataset was processed and transformed according to the following specifications: The Sex variable was encoded as a binary variable, with a value of 1 for males and 2 for females. Age was treated as a continuous variable, with values ranging from 18 to 90 years, represented as integer values. This range covers the distribution of ages in the study sample. BMI was calculated using the standard formula based on height and weight, resulting in continuous values ranging from 16.96 to 68.36. These values are represented as real numbers. PDF was processed based on family history using a PDF equation (Eq. 3.2), resulting in continuous decimal values ranging from 0 to 1. Fasting Blood sugar levels were recorded as continuous values, with values ranging from 55 to 250. Blood Pressure: Systolic blood pressure readings were recorded as continuous values, ranging from 60 to 100. Number of Pregnancies (for females): The number of pregnancies variable was encoded as a continuous variable ranging from 0 to 4.1. Diabetes Status (class): The Diabetes Status variable was encoded as a binary classification, where "Yes" indicates the presence of diabetes, and "No" indicates the absence of diabetes. Data Partitioning To develop reliable and generalizable machine learning models, it is essential to properly partition the available dataset into distinct subsets for different stages of model development and evaluation. This process helps prevent overfitting, ensures fair assessment of model performance, and optimizes model tuning procedures. The following section outlines the data partitioning strategy adopted in this study, including the types of data subsets, the data splitting techniques employed, and the specific proportions used to divide the dataset. Data Splitting and Methodology The dataset was partitioned into three subsets: Training Set: Used to train the models and learn the underlying patterns in the data. Validation Set: Used for hyperparameter tuning and model optimization to ensure generalization. Testing Set: Withheld from the training process and used solely for the final, unbiased evaluation of model performance. A stratified sampling technique was employed to partition the dataset, preserving the original class distribution in all subsets to reduce bias and improve evaluation accuracy. The data was split into a 60% (484 cases) training/validation set and a 40% (322 cases) hold-out test set. Within the training/validation set, model development and tuning utilized K-Fold Cross-Validation (K = 10). This method iteratively trains the model on 9 folds (90% of the training data). It validates it on the remaining 1-fold (10%), repeating this process 10 times until each fold has served as the validation set. This provides a robust estimate of model performance and generalization capability before final evaluation on the untouched test set (Lu et al., 2021 ; Wilimitis & Walsh, 2023). Ethical Consideration Written informed consent was obtained from each study participant before data collection. Moreover, a study was approved by the Ethics Committee (Standing Committee for Biosafety and Bioethics, NBC: 002.H-25.30). The protocol was previously published, and the study was conducted in accordance with the Helsinki Declaration. Results To evaluate the diagnostic performance of the machine learning models applied for diabetes prediction, confusion matrices were constructed for each algorithm. These matrices provided detailed insight into classification outcomes, including true positives (TP), true negatives (TN), false positives (FP), and false negatives (FN), enabling comparative analysis across algorithmic categories . Family history status Figure 2 illustrates that family history was prevalent among participants diagnosed with diabetes. A total of 56.82% of affected individuals reported having a father with diabetes, while 56.33% had a diabetic mother. Additionally, 52.36% indicated that a sibling was affected. In contrast, as shown in Figure 3, 64.02% of participants without diabetes reported no family history of the condition. The proportions of those with an affected father, mother, or sibling were notably lower in this group. Simple Computational Algorithms The Logistic Regression (LR) model, trained on 322 instances, achieved 160 TP and 150 TN, with 11 FP and 1 FN. Similarly, the Naive Bayes (NB) model yielded 159 TP, 148 TN, 13 FP, and 2 FN. Both models demonstrated high sensitivity, indicating strong capability in identifying diabetic cases, although their precision and specificity were comparatively lower due to moderate false-positive rates . Tree-Based Algorithms The Random Forest (RF) model recorded 155 TP and 159 TN, with only 2 FP and 6 FN, reflecting high specificity and overall classification accuracy. The Gradient Boosted Trees (GBT) model exhibited comparable performance, achieving 155 TP and 160 TN, with 1 FP and 6 FN. These ensemble-based models demonstrated superior balance between sensitivity and specificity, with minimal misclassification rates . Margin-Based Algorithms Support Vector Machines (SVM) produced 158 TP and 151 TN, alongside 10 FP and 3 FN, indicating balanced performance across all metrics. The Fast Large Margin (FLM) model yielded 155 TP and 145 TN, with a slightly elevated FP count of 16 and 6 FN, suggesting reduced specificity and precision relative to other models in this category . Performance Metrics Analysis As summarized in Table 2, the tree-based models outperformed others in overall accuracy, with GBT and RF achieving 97.8% and 97.5%, respectively, and the lowest classification error rates of 2.2% and 2.5%. GBT also attained the highest precision (99.4%), followed by RF (98.7%), reflecting minimal false-positive occurrences. In contrast, LR and NB recorded the highest sensitivity values (99.3% and 98.8%), underscoring their effectiveness in correctly identifying diabetic cases. Specificity was highest in GBT (99.4%) and RF (98.8%), indicating strong performance in correctly classifying non-diabetic individuals. The F-measure, representing the harmonic mean of precision and recall, ranged from 93.4% to 97.8%, with tree-based models demonstrating the most balanced and robust classification capabilities . Discussion Our study is a structured evaluation of machine learning (ML) algorithms for diabetes prediction in Libya, introducing a novel categorization framework that groups models into three distinct classes: simple computational algorithms (Logistic Regression and Naive Bayes), tree-based algorithms (Random Forest and Gradient Boosted Trees), and margin-based algorithms (Support Vector Machines and Fast Large Margin). This classification enhances methodological clarity and facilitates comparative analysis, aligning with recent literature that emphasizes algorithm selection based on clinical context (Alanazi, 2022 ; Ibrahim & Abdulazeez, 2021 ; Marnec, 2024 ). The study employed a rigorous 10-fold cross-validation approach, which strengthens the reliability of the results and mitigates overfitting—an essential consideration in medical prediction models (Wilimitis & Walsh, 2023). The findings indicate that family history is strongly associated with the development of diabetes, as the study revealed that more than half of the affected individuals had a parent or sibling with the disease. In contrast, the prevalence of a family history among non-affected individuals was relatively low, underscoring the role of genetic factors. Therefore, incorporating family history into screening and prevention programs is essential for early detection and effective disease control.(Hu et al., 2025 ; Smith et al., 2025 ). Among the evaluated models, tree-based algorithms demonstrated superior performance. Gradient Boosted Trees (GBT) achieved the highest accuracy (97.8%), precision (99.4%), specificity (99.4%), and F-measure (97.8%), followed closely by Random Forest (RF) with 97.5% accuracy and 98.7% precision. These models are particularly effective in capturing nonlinear relationships and complex feature interactions, making them well-suited for multifactorial diseases like diabetes (Lundberg et al., 2019 ; Parimbelli et al., 2023 ). In contrast, simple computational algorithms such as Logistic Regression (LR) and Naive Bayes (NB) excelled in sensitivity (99.3% and 98.8%, respectively), indicating strong detection of true diabetic cases. However, their lower specificity may result in higher false-positive rates, which could be problematic in clinical settings (Tan et al., 2024 ). Margin-based algorithms, including SVM and FLM, offered balanced performance but were generally outperformed by tree-based models in precision and specificity. The study’s findings significantly outperform prior research using public datasets like PIMA. For example, Khaleel & Al- Bakry. (2023) reported 94% accuracy for LR and 79% for NB, while Alzaboon et al. (2023) found RF and SVM achieving 76.3% and 73.9%, respectively (Khaleel & Al-Bakry, 2023 ; Alzboon et al., 2023(. In contrast, the current study—based on locally sourced Libyan data—achieved 96.2% (LR), 95.3% (NB), 97.5% (RF), and 96% (SVM), underscoring the importance of context-specific datasets in improving predictive performance. This localized approach not only enhances model accuracy but also supports the development of tailored clinical decision-support tools. Strategically, the proposed classification framework offers practical guidance for selecting optimal algorithms in clinical prediction tasks. It also opens pathways for hybrid modeling strategies that combine the high sensitivity of LR and NB with the high precision of RF and GBT. Such integration could lead to more balanced and effective diagnostic tools. These insights align with the findings of Kiran et al. ( 2025 ), whose 33-year review confirmed the dominance of tree-based ensemble methods in diabetes prediction (Kiran et al. ( 2025 ). Moreover, the study’s relevance extends beyond diabetes, echoing similar performance trends in cardiovascular disease and chronic kidney disease prediction, where tree-based models consistently outperform others (Ogunpola et al., 2024 ). Conclusion The study highlights the diagnostic superiority of tree-based algorithms, especially GBT and RF, for diabetes prediction in Libya. The categorization framework not only improves methodological transparency but also aids in developing hybrid models that balance sensitivity and specificity. These findings help enable more accurate, efficient, and context-aware clinical decision-making, emphasizing the importance of localized data and structured algorithm choices in public health research. Limitations While the study demonstrates strong predictive performance with machine learning models for diagnosing diabetes in Libya, several limitations should be considered. The dataset was exclusively based on Libyan clinical records, potentially limiting the applicability of the results to other populations. The study did not incorporate deep learning or hybrid ensemble models, which might improve accuracy. Moreover, the feature set was restricted to available clinical variables, omitting lifestyle and genetic factors that could affect diabetes risk. Future research should focus on increasing data diversity, testing more advanced algorithms, and validating models across a wider range of clinical settings. Declarations Ethics approval and consent to participate: In the material and method section. Acknowledgments The authors thank all the doctors in the healthcare centers and hospitals. We want to express our sincere gratitude to everyone who contributed to our research activities, including the data collectors Khadija Dahmani, Aida Alkhetoni, Asma Alshatewi, Yousef Almery, Salsabel Alyaquobi, Mohanned Abdulfethah, Mobasher Tarek, Zainab Said, and the Endocrine Department HTU team at Tripoli - Libya . Funding: This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors . Conflicts of interest: The authors declare that no competing interests exist . Availability of data and materials: Data is available in the Supplementary Material section . Contributions and Consent for publication: Manuscript writing: All authors; final approval of manuscript: All authors References Alanazi A (2022) Using machine learning for healthcare challenges and opportunities. In Informatics in Medicine Unlocked (Vol. 30). Elsevier Ltd. https://doi.org/10.1016/j.imu.2022.100924 Arkoudis N-A, Papadakos SP (2025a) Machine learning applications in healthcare clinical practice and research. World J Clin Cases 13(1). https://doi.org/10.12998/wjcc.v13.i1.99744 Corrao S, Janić M, Maggio V, Rizzo M (2025) Machine learning and deep learning in diabetology: revolutionizing diabetes care. In Frontiers in Clinical Diabetes and Healthcare . Front Media SA 6. https://doi.org/10.3389/fcdhc.2025.1547689 Davenport T, Kalakota R (2019) DIGITAL TECHNOLOGY The potential for artificial intelligence in healthcare. In Future Healthcare Journal (Vol. 6, Issue 2) Ghazizadeh Y (n.d.). Machine learning-based diabetes prediction: A comprehensive study on predictive modeling and risk assessment Journal of Clinical Images and Medical Case Reports. Open Access , 6 . https://doi.org/10.52768/2766-7820/3578 Holt RIG, Cockram CS, Ma RCW, Luk AOY (2024) Diabetes and infection: review of the epidemiology, mechanisms and principles of treatment. Diabetologia, vol 67. Springer Science and Business, pp 1168–1180. https://doi.org/10.1007/s00125-024-06102-x Media Deutschland GmbH Hu S, Ji W, Zhang Y, Zhu W, Sun H, Sun Y (2025) Risk factors for progression to type 2 diabetes in prediabetes: a systematic review and meta-analysis. BMC Public Health 25(1):1220. https://doi.org/10.1186/s12889-025-21404-4 Ibrahim IM, Abdulazeez AM (2021) The Role of Machine Learning Algorithms for Diagnosing Diseases. J Appl Sci Technol Trends 2(1):10–19. https://doi.org/10.38094/jastt20179 Islam R, Sultana A, Islam MR (2024) A comprehensive review for chronic disease prediction using machine learning algorithms. J Electr Syst Inform Technol 11(1). https://doi.org/10.1186/s43067-024-00150-4 Kanagarathinam K, Manikandan R, Kumar TS (2024) Machine learning algorithms-based decision support model for diabetes. Rev Comput Eng Res 11(1):16–29. https://doi.org/10.18488/76.v11i1.3598 Khaledi E, Ghomi F, Ghanei A, Meidany A, Shamsi F (2024) MACHINE LEARNING ENSEMBLE CLASSIFIERS TO PREDICT TYPE 2 DIABETES IN A COHORT STUDY. JP J Biostatistics 79–93. https://doi.org/10.17654/0973514325004 Khaleel A, Al-Bakry AM (2023) Diagnosis of diabetes using machine learning algorithms Show more Add to Mendeley Share Cite . https://doi.org/10.1016/j.matpr.2021.07.196Get Khalifa M, Albadawy M (2024) Artificial Intelligence for Clinical Prediction: Exploring Key Domains and Essential Functions. In Computer Methods and Programs in Biomedicine Update (Vol. 5). Elsevier B.V. https://doi.org/10.1016/j.cmpbup.2024.100148 Khosravi M, Zare Z, Mojtabaeian SM, Izadi R (2024) Artificial Intelligence and Decision-Making in Healthcare: A Thematic Analysis of a Systematic Review of Reviews. In Health Services Research and Managerial Epidemiology (Vol. 11). SAGE Publications Inc. https://doi.org/10.1177/23333928241234863 Kiran M, Xie Y, Anjum N, Ball G, Pierscionek B, Russell D (2025) Machine learning and artificial intelligence in type 2 diabetes prediction: a comprehensive 33-year bibliometric and literature analysis. In Frontiers in Digital Health . Front Media SA 7. https://doi.org/10.3389/fdgth.2025.1557467 Lai H, Huang H, Keshavjee K, Guergachi A, Gao X (2019) Predictive models for diabetes mellitus using machine learning techniques. BMC Endocr Disorders 19(1). https://doi.org/10.1186/s12902-019-0436-6 Liu T, Krentz A, Lu L, Curcin V (2025) Machine learning based prediction models for cardiovascular disease risk using electronic health records data: systematic review and meta-analysis. Eur Heart J - Digit Health 6(1):7–22. https://doi.org/10.1093/ehjdh/ztae080 Lu Y, Park Y, Chen L, Wang Y, De Sa C, Foster D (2021) Variance Reduced Training with Stratified Sampling for Forecasting Models Lundberg SM, Erion G, Chen H, DeGrave A, Prutkin JM, Nair B, Katz R, Himmelfarb J, Bansal N, Lee S-I (2019) Explainable AI for Trees: From Local Explanations to Global Understanding . http://arxiv.org/abs/1905.04610 Marnec (2024) SVM - Large Margin Classification | Machine Learning . https://marnec.github.io/ML/ML18/ Mowafaq S (2023) Alzboon, st, Mohammad Al-Batah, nd, Ahmad Abuashour, th, & Fuad Bader, A. A Comparative Study of Machine Learning Techniques for Early Prediction of Diabetes Ogunpola A, Saeed F, Basurra S, Albarrak AM, Qasem SN (2024) Machine Learning-Based Predictive Models for Detection of Cardiovascular Diseases. Diagnostics 14(2). https://doi.org/10.3390/diagnostics14020144 Oikonomou EK, Khera R (2023) Machine learning in precision diabetes care and cardiovascular risk prediction. In Cardiovascular Diabetology (Vol. 22, Issue 1). BioMed Central Ltd. https://doi.org/10.1186/s12933-023-01985-3 Ong, K. L., Stafford, L. K., McLaughlin, S. A., Boyko, E. J., Vollset, S. E., Smith,A. E., Dalton, B. E., Duprey, J., Cruz, J. A., Hagins, H., Lindstedt, P. A., Aali,A., Abate, Y. H., Abate, M. D., Abbasian, M., Abbasi-Kangevari, Z., Abbasi-Kangevari,M., Abd ElHafeez, S., Abd-Rabu, R., … Vos, T. (2023). Global, regional, and national burden of diabetes from 1990 to 2021, with projections of prevalence to 2050: a systematic analysis for the Global Burden of Disease Study 2021. The Lancet, 402(10397), 203–234. https://doi.org/10.1016/S0140-6736(23)01301-6 Parimbelli E, Buonocore TM, Nicora G, Michalowski W, Wilk S, Bellazzi R (2023) Why did AI get this one wrong? — Tree-based explanations of machine learning model predictions. Artificial Intelligence in Medicine , 135 . https://doi.org/10.1016/j.artmed.2022.102471 Pragyna Karmakar S, Debrupa Pal (2024) Artificial Intelligence. Int J Adv Res Sci Communication Technol 79–87. https://doi.org/10.48175/IJARSCT-19613 Sheikh H, Prins C, Schrijvers E (2023) Mission AI Research for Policy. Springer. https://link.springer.com/book/ 10.1007/978-3-031-21448-6 Smith KR, Meeks H, Curtis D, Brown BB, Kole K, Kowaleski-Jones L (2025) Family history of type 2 diabetes and the risk of type 2 diabetes among young and middle-aged adults. Chronic Dis Translational Med 11(1):46–56. https://doi.org/10.1002/cdt3.147 Tan Y, Sherwood B, Shenoy PP (2024) A naïve Bayes regularized logistic regression estimator for low-dimensional classification. International Journal of Approximate Reasoning , 172 . https://doi.org/10.1016/j.ijar.2024.109239 Ullah Z, Saleem F, Jamjoom M, Fakieh B, Kateb F, Ali AM, Shah B (2022) Detecting High-Risk Factors and Early Diagnosis of Diabetes Using Machine Learning Methods. Computational Intelligence and Neuroscience , 2022 . https://doi.org/10.1155/2022/2557795 Wang S, Chen Y, Cui Z, Lin L, Zong Y (2024) (n.d.). Diabetes Risk Analysis based on Machine Learning LASSO Regression Model. Www Centuryscipub Com 4. https://doi.org/10.53469/jtpes.2024.04(01).08 WHO (2023) Diabetes . https://www.who.int/news-room/fact-sheets/detail/diabetes Wilimitis D, Walsh CG (2023b) Practical Considerations and Applied Examples of Cross-Validation for Model Development and Evaluation in Health Care: Tutorial. JMIR AI 2(1). https://doi.org/10.2196/49023 Wu K (2024) Optimizing Diabetes Prediction with Machine Learning: Model Comparisons and Insights. J Sci Technol 5(4):41–51. https://doi.org/10.55662/jst.2024.5403 Tables Table .1: A summary of previous studies applying machine learning algorithms for diabetes prediction. Authors of the Study Used Algorithms Used Datasets Results Lai et al . (2019) LR, GBM, DT, RF Canadian CPCSSN, PIDD GBM achieved AROC 84.7%, Sensitivity 71.6%; LR AROC 84.0%, Sensitivity 73.4% Ullah et al . (2022) KNN, RF, XGBoost, Bagging and AdaBoost. BRFSS KNN achieved the highest accuracy: 98.38% Khaleel & Al-Bakry (2023) LR, NB, KNN PIDD LR accuracy 94%, NB 79%, KNN 69% Alzboon et al. (2023 ( LR, DT, RF, KNN, NB, SVM, Gradient Boosting, Neural Network PIDD Neural Network accuracy 78.57%, RF 76.30%, SVM 73.9% Wang et al . (2024) LASSO-regularized regression Alibaba Cloud Tianchi competition LASSO achieved AUC of 84.8% with ten-fold cross-validation Khaledi et al . (2025) LR, DT, RF, Bagging, AdaBoost, GBDT, SVM Shahedieh cohort AdaBoost achieved the highest accuracy: 86.2% Ghazizadeh et al. (2025) RF, SVM, LR, DT, NB, KNN PIDD RF highest accuracy: 84% , SVM and LR both 82% Kiran et al . (2025) Systematic Review Multiple studies (33-year review) Tree-based ensemble methods (Random Forest, Gradient Boosting) consistently outperformed others Table 2: The performance metrics of the machine learning built models for diabetes detection and prediction. Category Model Accuracy Error Rate Precision Recall F-Measure Specificity Simple Computational Algorithms LR 95.7% 4.3% 93.5% 98.1% 95.8% 93.2% NB 95.3% 4.7% 92.4% 98.8% 95.5% 92% Tree-based Algorithms RF 97.5% 2.5% 98.7% 96.3% 97.5% 98.8% GBT 97.8% 2.2% 99.4% 96.3% 97.8% 99.4% Margin-based Algorithms SVM 96% 4% 94.1% 98.1% 96.1% 93.8% FLM 93.2% 6.8% 90.6% 96.3% 93.4% 90.1% Additional Declarations No competing interests reported. Supplementary Files Data.xlsx Cite Share Download PDF Status: Posted Version 1 posted 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-7687597","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":519466248,"identity":"baea03c7-5a44-4c41-a966-036accad9d49","order_by":0,"name":"Sama Tarek Jadidi","email":"","orcid":"","institution":"National Center of Disease Control (NCDC)","correspondingAuthor":false,"prefix":"","firstName":"Sama","middleName":"Tarek","lastName":"Jadidi","suffix":""},{"id":519466249,"identity":"67716326-3524-415d-9b63-48ab3c6f04d2","order_by":1,"name":"Abdulbaset Mustafa Goweder","email":"","orcid":"","institution":"Libyan Academy","correspondingAuthor":false,"prefix":"","firstName":"Abdulbaset","middleName":"Mustafa","lastName":"Goweder","suffix":""},{"id":519466250,"identity":"2338da4a-b04e-4eb7-a63e-57d9508f23dc","order_by":2,"name":"Mohamed Hadi Mohamed Abdelhamid","email":"data:image/png;base64,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","orcid":"","institution":"National Center of Disease Control (NCDC)","correspondingAuthor":true,"prefix":"","firstName":"Mohamed","middleName":"Hadi Mohamed","lastName":"Abdelhamid","suffix":""}],"badges":[],"createdAt":"2025-09-23 01:54:44","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7687597/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7687597/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":92504395,"identity":"d928aa67-f6e9-4a9a-b643-ccd8f5b9cb85","added_by":"auto","created_at":"2025-09-30 12:18:30","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":62421,"visible":true,"origin":"","legend":"","description":"","filename":"Figure12.docx","url":"https://assets-eu.researchsquare.com/files/rs-7687597/v1/f27a5882c204b2065bc6e137.docx"},{"id":92505392,"identity":"c04eb711-24ba-4c64-8478-40f02dbcc547","added_by":"auto","created_at":"2025-09-30 12:26:30","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":321725,"visible":true,"origin":"","legend":"","description":"","filename":"AnEvaluationofMachineLearningCategories2209202512H40.docx","url":"https://assets-eu.researchsquare.com/files/rs-7687597/v1/0bd7d473c404247eb6e478fa.docx"},{"id":92504401,"identity":"1c24293e-11aa-4357-99b0-2467fcd4813c","added_by":"auto","created_at":"2025-09-30 12:18:30","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":19305,"visible":true,"origin":"","legend":"","description":"","filename":"Figure2.docx","url":"https://assets-eu.researchsquare.com/files/rs-7687597/v1/90d3295727684b84cb377309.docx"},{"id":92504399,"identity":"49aa9d41-37d0-4fc0-b355-b06eb8e8f466","added_by":"auto","created_at":"2025-09-30 12:18:30","extension":"docx","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":16865,"visible":true,"origin":"","legend":"","description":"","filename":"Table1.docx","url":"https://assets-eu.researchsquare.com/files/rs-7687597/v1/7ade4b4ed37ecf5805ad3f80.docx"},{"id":92505713,"identity":"19f7cf39-b186-448a-8021-3e1aff3819eb","added_by":"auto","created_at":"2025-09-30 12:34:30","extension":"docx","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":19265,"visible":true,"origin":"","legend":"","description":"","filename":"Figure3.docx","url":"https://assets-eu.researchsquare.com/files/rs-7687597/v1/897279565289c30fbcd4fdea.docx"},{"id":92504403,"identity":"038504ee-cdfc-44d5-a229-953c9f76eb61","added_by":"auto","created_at":"2025-09-30 12:18:30","extension":"docx","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":15423,"visible":true,"origin":"","legend":"","description":"","filename":"Table2.docx","url":"https://assets-eu.researchsquare.com/files/rs-7687597/v1/802e4aac8c9f7ec893774a55.docx"},{"id":92504402,"identity":"3c458d35-dff0-4d26-9bc3-7fd5945c3313","added_by":"auto","created_at":"2025-09-30 12:18:30","extension":"json","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":5747,"visible":true,"origin":"","legend":"","description":"","filename":"a39e3babfbe045f5bf01ea0f2fd9e552.json","url":"https://assets-eu.researchsquare.com/files/rs-7687597/v1/618722778a346bc680b2b79d.json"},{"id":92504408,"identity":"8ee0e057-16fb-4980-9c66-48cd4969f468","added_by":"auto","created_at":"2025-09-30 12:18:30","extension":"xlsx","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":135698,"visible":true,"origin":"","legend":"","description":"","filename":"Data.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-7687597/v1/92399e01d06ba10c8a69a9ea.xlsx"},{"id":92504410,"identity":"55decb6c-2887-44f2-96be-6735b364e792","added_by":"auto","created_at":"2025-09-30 12:18:30","extension":"xml","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":98769,"visible":true,"origin":"","legend":"","description":"","filename":"a39e3babfbe045f5bf01ea0f2fd9e5521enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7687597/v1/310ffb2e1aea6effdbc9910d.xml"},{"id":92505393,"identity":"f575a980-89d0-4312-a905-254697d79f74","added_by":"auto","created_at":"2025-09-30 12:26:30","extension":"png","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":40356,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7687597/v1/9cbf74f2fb7d35cabd1c991e.png"},{"id":92504407,"identity":"dea2bfd5-d797-42a5-bccc-dbbea50a35cf","added_by":"auto","created_at":"2025-09-30 12:18:30","extension":"xml","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":95263,"visible":true,"origin":"","legend":"","description":"","filename":"a39e3babfbe045f5bf01ea0f2fd9e5521structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7687597/v1/33c66ea38903e43adf01b8ed.xml"},{"id":92504409,"identity":"9f4dd83a-31df-499b-9d40-ed5c082dfa6b","added_by":"auto","created_at":"2025-09-30 12:18:30","extension":"html","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":106701,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7687597/v1/a1bab2ebf663382ffd5d2844.html"},{"id":92504396,"identity":"cb2ea7b4-f651-4dcf-9a57-0567802b82a7","added_by":"auto","created_at":"2025-09-30 12:18:30","extension":"jpeg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":159768,"visible":true,"origin":"","legend":"\u003cp\u003eExcluding particular data.\u003c/p\u003e","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7687597/v1/ba6bccc36cba52b172004344.jpeg"},{"id":92504398,"identity":"20fadaf9-980b-4f78-aba1-94155ac154de","added_by":"auto","created_at":"2025-09-30 12:18:30","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":21399,"visible":true,"origin":"","legend":"\u003cp\u003eDistribution of family history among participants diagnosed with diabetes.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-7687597/v1/4e00aed5470dc6e4632aa185.png"},{"id":92505391,"identity":"7a6b7064-740d-4e33-b14a-02b6bc4f69c5","added_by":"auto","created_at":"2025-09-30 12:26:30","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":21088,"visible":true,"origin":"","legend":"\u003cp\u003eDistribution of family history among participants diagnosed without diabetes.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-7687597/v1/f2f26666fe076c8ee93252cd.png"},{"id":92814985,"identity":"41196a2d-c2aa-40cf-a010-e3b7c41e8ef3","added_by":"auto","created_at":"2025-10-05 19:01:43","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1107152,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7687597/v1/bfd07b0c-b294-4ab9-afa2-d7d63071abd5.pdf"},{"id":92505712,"identity":"8d455975-9192-4a75-a013-8118e1772cf8","added_by":"auto","created_at":"2025-09-30 12:34:30","extension":"xlsx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":135698,"visible":true,"origin":"","legend":"","description":"","filename":"Data.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-7687597/v1/6b93a07c1b737150c761ddb2.xlsx"}],"financialInterests":"No competing interests reported.","formattedTitle":"An Evaluation of Machine Learning Categories for Diabetes Prediction and Detection in Libya: A Comparative Study","fulltext":[{"header":"Introduction","content":"\u003cp\u003eDiabetes mellitus is a chronic disease that is witnessing an increasing prevalence worldwide. This disease leads to serious complications, including cardiovascular diseases, kidney failure, retinopathy, neuropathy, and other complications that may result in limb amputation. Effective management and early prediction of diabetes are essential to reduce these risks and improve the quality of healthcare (Holt et al., \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Ong et al., 2023; WHO, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eOn the other hand, artificial Intelligence (AI) is a multidisciplinary branch of computer science that focuses on developing systems capable of simulating human intelligence and performing tasks requiring advanced cognitive abilities (Pragyna Karmakar et al., 2024; Sheikh et al., \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Significantly, AI has made substantial contributions to healthcare by enhancing clinical decision-making accuracy. Through its ability to analyze extensive datasets and identify complex patterns and relationships with high precision, AI empowers medical professionals to make more accurate and effective treatment decisions (Davenport \u0026amp; Kalakota, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Khosravi et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Moreover, in the healthcare domain, AI showcases immense potential in improving the diagnosis, prediction, and management of chronic diseases, including conditions such as diabetes, hypertension, heart disease, and kidney disease. By integrating AI into medical practices, healthcare providers can leverage its capabilities to optimize patient care and enhance overall health outcomes (Islam et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Khalifa \u0026amp; Albadawy, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Liu et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eIn fact, Machine learning (ML), a key area of artificial intelligence (AI), focuses on analyzing, classifying, and predicting outcomes based on patterns in complex datasets. In healthcare\u0026mdash;especially in diabetes management\u0026mdash;ML has emerged a transformative tool for early diagnosis, predicting disease progression, and optimizing treatment plans(Corrao et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Kanagarathinam et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Wu, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) by using ML techniques, various medical parameters such as blood glucose levels, blood pressure, body mass index (BMI), and genetic traits can be systematically examined to identify individuals at risk of developing diabetes. One of the most significant uses of ML in this field is its ability to detect early signs of the disease, enabling timely intervention and reducing the risk of complications (Arkoudis \u0026amp; Papadakos, 2025). Additionally, ML algorithms are used to predict diabetes-related complications, including diabetic retinopathy and cardiovascular issues, and to support personalized treatment plans by customizing recommendations for each patient, ultimately improving treatment results and reducing associated risks (Arkoudis \u0026amp; Papadakos, 2025; Oikonomou \u0026amp; Khera, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Several studies have shown that the prediction and diagnosis of diabetes, using techniques such as machine learning, deep learning, and neural networks, range from 69% to 98.38% in accuracy (Table .1).\u003c/p\u003e\u003cp\u003eLai et al. (\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2019\u003c/span\u003e) implemented a study to develop a predictive model for identifying individuals at risk of developing diabetes (Lai et al., \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). The study applied Logistic Regression and Gradient Boosting Machine models and compared their performance with Decision Tree and Random Forest models. In the same vein, the analysis relied on data from the Canadian CPCSSN database and the PIMA Indians dataset (PIDD). The results demonstrated that the GBM model outperformed other models with an AROC of 84.7% and a sensitivity of 71.6%, while LR achieved an AROC of 84.0% and a sensitivity of 73.4%.\u003c/p\u003e\u003cp\u003eFurthermore, Ullah et al., (\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) conducted a study to detect diabetes using machine learning techniques (Ullah et al., \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). The applied algorithms included K-Nearest Neighbors (KNN), Random Forest, XGBoost, Bagging, and AdaBoost, alongside the SMOTE-ENN technique to address data imbalance. In fact, the study used data from the publicly available Behavioral Risk Factor Surveillance System (BRFSS) database. Importantly, the findings demonstrated that the KNN algorithm achieved the highest accuracy at 98.38%.\u003c/p\u003e\u003cp\u003eLikewise, Khaleel \u0026amp; Al-bakry (\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) showed that to develop a model capable of predicting diabetes diagnosis Click or tap here to enter text.(Khaleel \u0026amp; Al-Bakry, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Several machine learning algorithms were applied, including Logistic Regression, Naive Bayes, and K-Nearest Neighbors, using the Pima Indian Diabetes Database (PIDD). Their comparative results revealed that LR achieved the highest accuracy at 94%, followed by NB at 79%, and KNN at 69%.\u003c/p\u003e\u003cp\u003eSimilarly, Alzboon et al., (2023) evaluated the effectiveness of several machine learning algorithms for the early prediction of diabetes (Alzboon et al., 2023). The algorithms included Logistic Regression, Decision Tree, Random Forest, K-Nearest Neighbors, Naive Bayes, Support Vector Machine, Gradient Boosting, and a neural network. The analysis was conducted using the Pima Indian Diabetes Database. According to their findings, the neural network algorithm outperformed others, achieving an accuracy of 78.57%, followed by Random Forest at 76.30%, and Support Vector Machine at 73.9%.\u003c/p\u003e\u003cp\u003eMoreover, Wang et al. (\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) carried out a study focusing on predicting diabetes risk using a LASSO-regularized regression model, based on data from the Alibaba Cloud Tianchi competition (Wang et al., \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). The results indicated that the LASSO model achieved an AUC of 84.8%, employing a ten-fold cross-validation approach.\u003c/p\u003e\u003cp\u003eIn a recent study, Khaledi \u003cem\u003eet al.\u003c/em\u003e (2025) conducted a study aimed at developing machine learning models to predict diabetes. In this context, the study utilized Logistic Regression, Decision Tree, Random Forest, bagging methods, boosting techniques (including AdaBoost and Gradient Boosted Decision Trees (GBDT)), and Support Vector Machine (SVM) (Khaledi et al., 2025). The data for model development were obtained from the Shahedieh cohort. Notably, the findings revealed that the AdaBoost model achieved the highest accuracy, reaching 86.2%.\u003c/p\u003e\u003cp\u003eIn a similar vein, Ghazizadeh (2025) evaluated the performance of various machine learning models for diabetes prediction using the PIMA Indians dataset Ghazizadeh 2025). The models included Random Forest, Support Vector Machine, Logistic Regression, Decision Tree, Naive Bayes, and K-Nearest Neighbors. According to their comparative analysis, the Random Forest model achieved the highest accuracy at 84%, while both Support Vector Machines and Logistic Regression models recorded an accuracy of 82%.\u003c/p\u003e\u003cp\u003eFurthermore, Kiran et al. (\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) conducted a comprehensive systematic review of literature spanning 33 years on diabetes prediction using machine learning techniques (Kiran et al., \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). The review indicated that Decision Tree, Random Forest, Gradient Boosting Machines (GBMs), and Neural Networks (NNs) consistently demonstrated strong predictive performance. Furthermore, it highlighted that Tree-based ensemble methods, particularly Random Forest and Gradient Boosting, often outperformed other algorithms in terms of prediction accuracy.\u003c/p\u003e\u003cp\u003eIn this context, this study aims to detect and predict DM using Simple Computational, Tree-based, and Margin-based algorithms. It compares the performance of the built models to determine which category of algorithms is most effective for the task of DM detection and prediction.\u003c/p\u003e"},{"header":"Materials and Methods","content":"\u003cp\u003eIn this research study, six different machine learning algorithms were adopted and applied to build different models for the detection and prediction of diabetes among Libyan patients. These machine learning algorithms belong to three main categories: Simple Computational algorithms, Tree-based algorithms, and Margin-based algorithms. The built models are evaluated to assess their performance in order to compare the three main categories.\u003c/p\u003e\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003eStudy Population\u003c/h2\u003e\u003cp\u003eIn this study, the dataset was systematically collected, comprising a total of 806 patients classified into two groups: 403 non-diabetic individuals (control group) and 403 diabetic individuals (case group). This balanced distribution ensures better performance in predicting the probability of diabetes.\u003c/p\u003e\u003cp\u003eTo improve the accuracy of the analysis and ensure more reliable predictions, the selection of features for this study was based on previous literature, supplemented by consultations with endocrinologists, diabetes specialists, and physicians from other specialties. Subsequently, a set of features deemed essential and highly impactful for diabetes prediction was selected. The features included in this study comprised age (restricted to adults), sex, and the number of pregnancies for female participants. Additionally, the study incorporated the measurement of Body Mass Index (BMI), which was calculated based on height and weight data. Clinical indicators such as fasting blood glucose level and diastolic blood pressure were also included, given their well-established significance in assessing the risk of developing diabetes. Furthermore, the Pedigree Diabetes Function (PDF) was incorporated, which accounts for genetic factors by determining whether a participant has a parent or sibling diagnosed with diabetes. Finally, the dataset included the diabetes status of each participant, which served as the target variable for the prediction process.\u003c/p\u003e\u003cp\u003eIn accordance with the features adopted in this study, the distribution of sex features among participants was determined, with 359 males and 447 females. This distribution reflects a roughly balanced representation between the two groups, thereby attempting to enhance the accuracy of the study's results.\u003c/p\u003e\u003cp\u003eAll records related to individuals under the age of 18 were excluded to ensure consideration only of adults. Additionally, any data obtained from patient files were removed to ensure the accuracy and reliability of the dataset (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eData Collection and Sources\u003c/h3\u003e\n\u003cp\u003eThe data were gathered by a team of 12 healthcare professionals (Tripoli Teaching Hospital, National Center for Diabetes and Endocrinology, National Center for Disease Control - Souk Jomaa Branch, Janzour Kidney Dialysis Services Center, Qasr Alhamra Pharmacy, and Al-Qasr Pharmacy). professionals, including Physicians, Nurses, Pharmacists, and Medical Students. Data collection started in February 2023 and proceeded until August 2024. The goal was to gather the necessary number of data points while ensuring their accuracy and reliability. Additionally, efforts were made to balance the sample distribution to accurately represent the targeted groups. Moreover, the Kobo Toolbox software was used for data collection (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://ee.kobotoolbox.org/x/nIj5eTMb\u003c/span\u003e\u003cspan address=\"https://ee.kobotoolbox.org/x/nIj5eTMb\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e), as it is an effective tool for gathering medical field data. It allows data collection using mobile devices, which significantly facilitates data management. After the data was collected, it was stored within the Kobo software. Subsequently, the data was transferred to Microsoft Excel, where basic calculations were applied to ensure accurate analysis and further processing.\u003c/p\u003e\n\u003ch3\u003eStudy Questionnaire\u003c/h3\u003e\n\u003cp\u003eThe questionnaire began with a screening question to identify participants with a prior diabetes diagnosis. It then consisted of three thematic sections: the first collected demographic and personal information (such as age, sex, marital status, number of children, nationality, and occupation) to provide context for risk factors; the second gathered key health indicators (height, weight, fasting blood glucose, diastolic blood pressure) to explore biological causes; and the third examined family history by noting diabetes in close relatives and the number of affected siblings to evaluate hereditary and genetic influences.\u003c/p\u003e\n\u003ch3\u003eData Cleaning and Calculation Methods\u003c/h3\u003e\n\u003cp\u003eData cleaning is considered one of the essential steps to ensure the accuracy of the results. Missing data were addressed by deleting records containing any incomplete information. Similarly, outliers or errors were handled in the same manner by removing the entire record. Furthermore, data formatting was verified to ensure consistency across different measurement units. In addition to the cleaning process, computational methods were applied to calculate the BMI and PDF indicators: The BMI was calculated using the standard formula:\u003c/p\u003e\u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:BMI=\\frac{Weight\\left(kg\\right)}{{height\\left(m\\right)}^{2}}\\)\u003c/span\u003e\u003c/span\u003e \u003cem\u003eEq.\u0026nbsp;(1\u003c/em\u003e)\u003c/p\u003e\u003cp\u003eThe Pedigree Diabetes Function (PDF) is a risk factor derived from family history and was calculated using the following formula:\u003c/p\u003e\u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:PDF=\\left(0.3*\\:Siblings\\right)\\:+\\:\\left(0.4\\:*\\:Mother\\right)\\:+\\:\\left(0.3\\:*\\:Father\\right)\\:\\)\u003c/span\u003e\u003c/span\u003e \u003cem\u003eEq.\u0026nbsp;(2\u003c/em\u003e)\u003c/p\u003e\u003cp\u003eTable\u0026nbsp;1 shows the values to be substituted for the variables (Siblings, Mother, and Father) of Eq.\u0026nbsp;(2).\u003c/p\u003e\u003cp\u003eThe collected dataset was processed and transformed according to the following specifications:\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003eThe Sex variable was encoded as a binary variable, with a value of 1 for males and 2 for females.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eAge was treated as a continuous variable, with values ranging from 18 to 90 years, represented as integer values. This range covers the distribution of ages in the study sample.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eBMI was calculated using the standard formula based on height and weight, resulting in continuous values ranging from 16.96 to 68.36. These values are represented as real numbers.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003ePDF was processed based on family history using a PDF equation (Eq.\u0026nbsp;3.2), resulting in continuous decimal values ranging from 0 to 1.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eFasting Blood sugar levels were recorded as continuous values, with values ranging from 55 to 250.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eBlood Pressure: Systolic blood pressure readings were recorded as continuous values, ranging from 60 to 100.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eNumber of Pregnancies (for females): The number of pregnancies variable was encoded as a continuous variable ranging from 0 to 4.1.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eDiabetes Status (class): The Diabetes Status variable was encoded as a binary classification, where \"Yes\" indicates the presence of diabetes, and \"No\" indicates the absence of diabetes.\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e\n\u003ch3\u003eData Partitioning\u003c/h3\u003e\n\u003cp\u003eTo develop reliable and generalizable machine learning models, it is essential to properly partition the available dataset into distinct subsets for different stages of model development and evaluation. This process helps prevent overfitting, ensures fair assessment of model performance, and optimizes model tuning procedures. The following section outlines the data partitioning strategy adopted in this study, including the types of data subsets, the data splitting techniques employed, and the specific proportions used to divide the dataset.\u003c/p\u003e\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\u003ch2\u003eData Splitting and Methodology\u003c/h2\u003e\u003cp\u003eThe dataset was partitioned into three subsets:\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003eTraining Set: Used to train the models and learn the underlying patterns in the data.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eValidation Set: Used for hyperparameter tuning and model optimization to ensure generalization.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eTesting Set: Withheld from the training process and used solely for the final, unbiased evaluation of model performance.\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e\u003cp\u003eA stratified sampling technique was employed to partition the dataset, preserving the original class distribution in all subsets to reduce bias and improve evaluation accuracy. The data was split into a 60% (484 cases) training/validation set and a 40% (322 cases) hold-out test set. Within the training/validation set, model development and tuning utilized K-Fold Cross-Validation (K\u0026thinsp;=\u0026thinsp;10). This method iteratively trains the model on 9 folds (90% of the training data). It validates it on the remaining 1-fold (10%), repeating this process 10 times until each fold has served as the validation set. This provides a robust estimate of model performance and generalization capability before final evaluation on the untouched test set (Lu et al., \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Wilimitis \u0026amp; Walsh, 2023).\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eEthical Consideration\u003c/h3\u003e\n\u003cp\u003e Written informed consent was obtained from each study participant before data collection. Moreover, a study was approved by the Ethics Committee (Standing Committee for Biosafety and Bioethics, NBC: 002.H-25.30). The protocol was previously published, and the study was conducted in accordance with the Helsinki Declaration.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eTo evaluate the diagnostic performance of the machine learning models applied for diabetes prediction, confusion matrices were constructed for each algorithm. These matrices provided detailed insight into classification outcomes, including true positives (TP), true negatives (TN), false positives (FP), and false negatives (FN), enabling comparative analysis across algorithmic categories\u003cspan dir=\"RTL\"\u003e.\u003c/span\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFamily history status\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFigure 2 illustrates that family history was prevalent among participants diagnosed with diabetes. A total of 56.82% of affected individuals reported having a father with diabetes, while 56.33% had a diabetic mother. Additionally, 52.36% indicated that a sibling was affected. In contrast, as shown in Figure 3, 64.02% of participants without diabetes reported no family history of the condition. The proportions of those with an affected father, mother, or sibling were notably lower\u0026nbsp;in\u0026nbsp;this\u0026nbsp;group.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSimple Computational Algorithms\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe Logistic Regression (LR) model, trained on 322 instances, achieved 160 TP and 150 TN, with 11 FP and 1 FN. Similarly, the Naive Bayes (NB) model yielded 159 TP, 148 TN, 13 FP, and 2 FN. Both models demonstrated high sensitivity, indicating strong capability in identifying diabetic cases, although their precision and specificity were comparatively lower due to moderate false-positive rates\u003cspan dir=\"RTL\"\u003e.\u003c/span\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTree-Based Algorithms\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe Random Forest (RF) model recorded 155 TP and 159 TN, with only 2 FP and 6 FN, reflecting high specificity and overall classification accuracy. The Gradient Boosted Trees (GBT) model exhibited comparable performance, achieving 155 TP and 160 TN, with 1 FP and 6 FN. These ensemble-based models demonstrated superior balance between sensitivity and specificity, with minimal misclassification rates\u003cspan dir=\"RTL\"\u003e.\u003c/span\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMargin-Based Algorithms\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSupport Vector Machines (SVM) produced 158 TP and 151 TN, alongside 10 FP and 3 FN, indicating balanced performance across all metrics. The Fast Large Margin (FLM) model yielded 155 TP and 145 TN, with a slightly elevated FP count of 16 and 6 FN, suggesting reduced specificity and precision relative to other models in this category\u003cspan dir=\"RTL\"\u003e.\u003c/span\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePerformance Metrics Analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAs summarized in Table 2, the tree-based models outperformed others in overall accuracy, with GBT and RF achieving 97.8% and 97.5%, respectively, and the lowest classification error rates of 2.2% and 2.5%. GBT also attained the highest precision (99.4%), followed by RF (98.7%), reflecting minimal false-positive occurrences. In contrast, LR and NB recorded the highest sensitivity values (99.3% and 98.8%), underscoring their effectiveness in correctly identifying diabetic cases. Specificity was highest in GBT (99.4%) and RF (98.8%), indicating strong performance in correctly classifying non-diabetic individuals. The F-measure, representing the harmonic mean of precision and recall, ranged from 93.4% to 97.8%, with tree-based models demonstrating the most balanced and robust classification capabilities\u003cstrong\u003e.\u003c/strong\u003e\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eOur study is a structured evaluation of machine learning (ML) algorithms for diabetes prediction in Libya, introducing a novel categorization framework that groups models into three distinct classes: simple computational algorithms (Logistic Regression and Naive Bayes), tree-based algorithms (Random Forest and Gradient Boosted Trees), and margin-based algorithms (Support Vector Machines and Fast Large Margin). This classification enhances methodological clarity and facilitates comparative analysis, aligning with recent literature that emphasizes algorithm selection based on clinical context (Alanazi, \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Ibrahim \u0026amp; Abdulazeez, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Marnec, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). The study employed a rigorous 10-fold cross-validation approach, which strengthens the reliability of the results and mitigates overfitting\u0026mdash;an essential consideration in medical prediction models (Wilimitis \u0026amp; Walsh, 2023).\u003c/p\u003e\u003cp\u003eThe findings indicate that family history is strongly associated with the development of diabetes, as the study revealed that more than half of the affected individuals had a parent or sibling with the disease. In contrast, the prevalence of a family history among non-affected individuals was relatively low, underscoring the role of genetic factors. Therefore, incorporating family history into screening and prevention programs is essential for early detection and effective disease control.(Hu et al., \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Smith et al., \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eAmong the evaluated models, tree-based algorithms demonstrated superior performance. Gradient Boosted Trees (GBT) achieved the highest accuracy (97.8%), precision (99.4%), specificity (99.4%), and F-measure (97.8%), followed closely by Random Forest (RF) with 97.5% accuracy and 98.7% precision. These models are particularly effective in capturing nonlinear relationships and complex feature interactions, making them well-suited for multifactorial diseases like diabetes (Lundberg et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Parimbelli et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). In contrast, simple computational algorithms such as Logistic Regression (LR) and Naive Bayes (NB) excelled in sensitivity (99.3% and 98.8%, respectively), indicating strong detection of true diabetic cases. However, their lower specificity may result in higher false-positive rates, which could be problematic in clinical settings (Tan et al., \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Margin-based algorithms, including SVM and FLM, offered balanced performance but were generally outperformed by tree-based models in precision and specificity.\u003c/p\u003e\u003cp\u003eThe study\u0026rsquo;s findings significantly outperform prior research using public datasets like PIMA. For example, Khaleel \u0026amp; Al- Bakry. (2023) reported 94% accuracy for LR and 79% for NB, while Alzaboon et al. (2023) found RF and SVM achieving 76.3% and 73.9%, respectively (Khaleel \u0026amp; Al-Bakry, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Alzboon et al., 2023(. In contrast, the current study\u0026mdash;based on locally sourced Libyan data\u0026mdash;achieved 96.2% (LR), 95.3% (NB), 97.5% (RF), and 96% (SVM), underscoring the importance of context-specific datasets in improving predictive performance. This localized approach not only enhances model accuracy but also supports the development of tailored clinical decision-support tools.\u003c/p\u003e\u003cp\u003eStrategically, the proposed classification framework offers practical guidance for selecting optimal algorithms in clinical prediction tasks. It also opens pathways for hybrid modeling strategies that combine the high sensitivity of LR and NB with the high precision of RF and GBT. Such integration could lead to more balanced and effective diagnostic tools. These insights align with the findings of Kiran et al. (\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2025\u003c/span\u003e), whose 33-year review confirmed the dominance of tree-based ensemble methods in diabetes prediction (Kiran et al. (\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Moreover, the study\u0026rsquo;s relevance extends beyond diabetes, echoing similar performance trends in cardiovascular disease and chronic kidney disease prediction, where tree-based models consistently outperform others (Ogunpola et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThe study highlights the diagnostic superiority of tree-based algorithms, especially GBT and RF, for diabetes prediction in Libya. The categorization framework not only improves methodological transparency but also aids in developing hybrid models that balance sensitivity and specificity. These findings help enable more accurate, efficient, and context-aware clinical decision-making, emphasizing the importance of localized data and structured algorithm choices in public health research.\u003c/p\u003e\u003cdiv id=\"Sec18\" class=\"Section2\"\u003e\u003ch2\u003eLimitations\u003c/h2\u003e\u003cp\u003eWhile the study demonstrates strong predictive performance with machine learning models for diagnosing diabetes in Libya, several limitations should be considered. The dataset was exclusively based on Libyan clinical records, potentially limiting the applicability of the results to other populations. The study did not incorporate deep learning or hybrid ensemble models, which might improve accuracy. Moreover, the feature set was restricted to available clinical variables, omitting lifestyle and genetic factors that could affect diabetes risk. Future research should focus on increasing data diversity, testing more advanced algorithms, and validating models across a wider range of clinical settings.\u003c/p\u003e\u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003eEthics approval and consent to participate: In the material and method section.\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgments\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors thank all the doctors in the healthcare centers and hospitals. We want to express our sincere gratitude to everyone who contributed to our research activities, including the data collectors \u003cstrong\u003eKhadija Dahmani, Aida Alkhetoni, Asma Alshatewi, Yousef Almery, Salsabel Alyaquobi, Mohanned Abdulfethah, Mobasher Tarek, Zainab Said,\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003eand the\u003cstrong\u003e\u0026nbsp;\u003cstrong\u003eEndocrine Department HTU team at Tripoli - Libya\u003c/strong\u003e\u003c/strong\u003e.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding:\u003c/strong\u003e This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors\u003cspan dir=\"RTL\"\u003e.\u003c/span\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflicts of interest:\u003c/strong\u003e The authors declare that no competing interests exist\u003cspan dir=\"RTL\"\u003e.\u003c/span\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials:\u003c/strong\u003e Data is available in the Supplementary Material section\u003cspan dir=\"RTL\"\u003e.\u003c/span\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eContributions and Consent for publication:\u003c/strong\u003e Manuscript writing: All authors; final approval of manuscript: All authors\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAlanazi A (2022) Using machine learning for healthcare challenges and opportunities. In \u003cem\u003eInformatics in Medicine Unlocked\u003c/em\u003e (Vol. 30). Elsevier Ltd. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.imu.2022.100924\u003c/span\u003e\u003cspan address=\"10.1016/j.imu.2022.100924\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eArkoudis N-A, Papadakos SP (2025a) Machine learning applications in healthcare clinical practice and research. World J Clin Cases 13(1). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.12998/wjcc.v13.i1.99744\u003c/span\u003e\u003cspan address=\"10.12998/wjcc.v13.i1.99744\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCorrao S, Janić M, Maggio V, Rizzo M (2025) Machine learning and deep learning in diabetology: revolutionizing diabetes care. In \u003cem\u003eFrontiers in Clinical Diabetes and Healthcare\u003c/em\u003e. Front Media SA 6. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fcdhc.2025.1547689\u003c/span\u003e\u003cspan address=\"10.3389/fcdhc.2025.1547689\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDavenport T, Kalakota R (2019) DIGITAL TECHNOLOGY The potential for artificial intelligence in healthcare. In \u003cem\u003eFuture Healthcare Journal\u003c/em\u003e (Vol. 6, Issue 2)\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGhazizadeh Y (n.d.). Machine learning-based diabetes prediction: A comprehensive study on predictive modeling and risk assessment Journal of Clinical Images and Medical Case Reports. \u003cem\u003eOpen Access\u003c/em\u003e, \u003cem\u003e6\u003c/em\u003e. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.52768/2766-7820/3578\u003c/span\u003e\u003cspan address=\"10.52768/2766-7820/3578\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHolt RIG, Cockram CS, Ma RCW, Luk AOY (2024) Diabetes and infection: review of the epidemiology, mechanisms and principles of treatment. Diabetologia, vol 67. Springer Science and Business, pp 1168\u0026ndash;1180. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s00125-024-06102-x\u003c/span\u003e\u003cspan address=\"10.1007/s00125-024-06102-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003eMedia Deutschland GmbH\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHu S, Ji W, Zhang Y, Zhu W, Sun H, Sun Y (2025) Risk factors for progression to type 2 diabetes in prediabetes: a systematic review and meta-analysis. BMC Public Health 25(1):1220. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12889-025-21404-4\u003c/span\u003e\u003cspan address=\"10.1186/s12889-025-21404-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eIbrahim IM, Abdulazeez AM (2021) The Role of Machine Learning Algorithms for Diagnosing Diseases. J Appl Sci Technol Trends 2(1):10\u0026ndash;19. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.38094/jastt20179\u003c/span\u003e\u003cspan address=\"10.38094/jastt20179\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eIslam R, Sultana A, Islam MR (2024) A comprehensive review for chronic disease prediction using machine learning algorithms. J Electr Syst Inform Technol 11(1). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s43067-024-00150-4\u003c/span\u003e\u003cspan address=\"10.1186/s43067-024-00150-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKanagarathinam K, Manikandan R, Kumar TS (2024) Machine learning algorithms-based decision support model for diabetes. Rev Comput Eng Res 11(1):16\u0026ndash;29. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.18488/76.v11i1.3598\u003c/span\u003e\u003cspan address=\"10.18488/76.v11i1.3598\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKhaledi E, Ghomi F, Ghanei A, Meidany A, Shamsi F (2024) MACHINE LEARNING ENSEMBLE CLASSIFIERS TO PREDICT TYPE 2 DIABETES IN A COHORT STUDY. JP J Biostatistics 79\u0026ndash;93. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.17654/0973514325004\u003c/span\u003e\u003cspan address=\"10.17654/0973514325004\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKhaleel A, Al-Bakry AM (2023) \u003cem\u003eDiagnosis of diabetes using machine learning algorithms Show more Add to Mendeley Share Cite\u003c/em\u003e. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.matpr.2021.07.196Get\u003c/span\u003e\u003cspan address=\"10.1016/j.matpr.2021.07.196Get\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKhalifa M, Albadawy M (2024) Artificial Intelligence for Clinical Prediction: Exploring Key Domains and Essential Functions. In \u003cem\u003eComputer Methods and Programs in Biomedicine Update\u003c/em\u003e (Vol. 5). Elsevier B.V. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.cmpbup.2024.100148\u003c/span\u003e\u003cspan address=\"10.1016/j.cmpbup.2024.100148\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKhosravi M, Zare Z, Mojtabaeian SM, Izadi R (2024) Artificial Intelligence and Decision-Making in Healthcare: A Thematic Analysis of a Systematic Review of Reviews. In \u003cem\u003eHealth Services Research and Managerial Epidemiology\u003c/em\u003e (Vol. 11). SAGE Publications Inc. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/23333928241234863\u003c/span\u003e\u003cspan address=\"10.1177/23333928241234863\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKiran M, Xie Y, Anjum N, Ball G, Pierscionek B, Russell D (2025) Machine learning and artificial intelligence in type 2 diabetes prediction: a comprehensive 33-year bibliometric and literature analysis. In \u003cem\u003eFrontiers in Digital Health\u003c/em\u003e. Front Media SA 7. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fdgth.2025.1557467\u003c/span\u003e\u003cspan address=\"10.3389/fdgth.2025.1557467\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLai H, Huang H, Keshavjee K, Guergachi A, Gao X (2019) Predictive models for diabetes mellitus using machine learning techniques. BMC Endocr Disorders 19(1). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12902-019-0436-6\u003c/span\u003e\u003cspan address=\"10.1186/s12902-019-0436-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLiu T, Krentz A, Lu L, Curcin V (2025) Machine learning based prediction models for cardiovascular disease risk using electronic health records data: systematic review and meta-analysis. Eur Heart J - Digit Health 6(1):7\u0026ndash;22. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/ehjdh/ztae080\u003c/span\u003e\u003cspan address=\"10.1093/ehjdh/ztae080\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLu Y, Park Y, Chen L, Wang Y, De Sa C, Foster D (2021) \u003cem\u003eVariance Reduced Training with Stratified Sampling for Forecasting Models\u003c/em\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLundberg SM, Erion G, Chen H, DeGrave A, Prutkin JM, Nair B, Katz R, Himmelfarb J, Bansal N, Lee S-I (2019) \u003cem\u003eExplainable AI for Trees: From Local Explanations to Global Understanding\u003c/em\u003e. http://arxiv.org/abs/1905.04610\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMarnec (2024) \u003cem\u003eSVM - Large Margin Classification | Machine Learning\u003c/em\u003e. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://marnec.github.io/ML/ML18/\u003c/span\u003e\u003cspan address=\"https://marnec.github.io/ML/ML18/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMowafaq S (2023) Alzboon, st, Mohammad Al-Batah, nd, Ahmad Abuashour, th, \u0026amp; Fuad Bader, A. \u003cem\u003eA Comparative Study of Machine Learning Techniques for Early Prediction of Diabetes\u003c/em\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eOgunpola A, Saeed F, Basurra S, Albarrak AM, Qasem SN (2024) Machine Learning-Based Predictive Models for Detection of Cardiovascular Diseases. Diagnostics 14(2). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/diagnostics14020144\u003c/span\u003e\u003cspan address=\"10.3390/diagnostics14020144\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eOikonomou EK, Khera R (2023) Machine learning in precision diabetes care and cardiovascular risk prediction. In \u003cem\u003eCardiovascular Diabetology\u003c/em\u003e (Vol. 22, Issue 1). BioMed Central Ltd. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12933-023-01985-3\u003c/span\u003e\u003cspan address=\"10.1186/s12933-023-01985-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eOng, K. L., Stafford, L. K., McLaughlin, S. A., Boyko, E. J., Vollset, S. E., Smith,A. E., Dalton, B. E., Duprey, J., Cruz, J. A., Hagins, H., Lindstedt, P. A., Aali,A., Abate, Y. H., Abate, M. D., Abbasian, M., Abbasi-Kangevari, Z., Abbasi-Kangevari,M., Abd ElHafeez, S., Abd-Rabu, R., \u0026hellip; Vos, T. (2023). Global, regional, and national burden of diabetes from 1990 to 2021, with projections of prevalence to 2050: a systematic analysis for the Global Burden of Disease Study 2021. The Lancet, 402(10397), 203\u0026ndash;234. https://doi.org/10.1016/S0140-6736(23)01301-6\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eParimbelli E, Buonocore TM, Nicora G, Michalowski W, Wilk S, Bellazzi R (2023) Why did AI get this one wrong? \u0026mdash; Tree-based explanations of machine learning model predictions. \u003cem\u003eArtificial Intelligence in Medicine\u003c/em\u003e, \u003cem\u003e135\u003c/em\u003e. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.artmed.2022.102471\u003c/span\u003e\u003cspan address=\"10.1016/j.artmed.2022.102471\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePragyna Karmakar S, Debrupa Pal (2024) Artificial Intelligence. Int J Adv Res Sci Communication Technol 79\u0026ndash;87. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.48175/IJARSCT-19613\u003c/span\u003e\u003cspan address=\"10.48175/IJARSCT-19613\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSheikh H, Prins C, Schrijvers E (2023) Mission AI Research for Policy. Springer. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://link.springer.com/book/\u003c/span\u003e\u003cspan address=\"https://link.springer.com/book/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/978-3-031-21448-6\u003c/span\u003e\u003cspan address=\"10.1007/978-3-031-21448-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSmith KR, Meeks H, Curtis D, Brown BB, Kole K, Kowaleski-Jones L (2025) Family history of type 2 diabetes and the risk of type 2 diabetes among young and middle-aged adults. Chronic Dis Translational Med 11(1):46\u0026ndash;56. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/cdt3.147\u003c/span\u003e\u003cspan address=\"10.1002/cdt3.147\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTan Y, Sherwood B, Shenoy PP (2024) A na\u0026iuml;ve Bayes regularized logistic regression estimator for low-dimensional classification. \u003cem\u003eInternational Journal of Approximate Reasoning\u003c/em\u003e, \u003cem\u003e172\u003c/em\u003e. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ijar.2024.109239\u003c/span\u003e\u003cspan address=\"10.1016/j.ijar.2024.109239\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eUllah Z, Saleem F, Jamjoom M, Fakieh B, Kateb F, Ali AM, Shah B (2022) Detecting High-Risk Factors and Early Diagnosis of Diabetes Using Machine Learning Methods. \u003cem\u003eComputational Intelligence and Neuroscience\u003c/em\u003e, \u003cem\u003e2022\u003c/em\u003e. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1155/2022/2557795\u003c/span\u003e\u003cspan address=\"10.1155/2022/2557795\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang S, Chen Y, Cui Z, Lin L, Zong Y (2024) (n.d.). Diabetes Risk Analysis based on Machine Learning LASSO Regression Model. Www Centuryscipub Com 4. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.53469/jtpes.2024.04(01).08\u003c/span\u003e\u003cspan address=\"10.53469/jtpes.2024.04(01).08\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWHO (2023) \u003cem\u003eDiabetes\u003c/em\u003e. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.who.int/news-room/fact-sheets/detail/diabetes\u003c/span\u003e\u003cspan address=\"https://www.who.int/news-room/fact-sheets/detail/diabetes\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWilimitis D, Walsh CG (2023b) Practical Considerations and Applied Examples of Cross-Validation for Model Development and Evaluation in Health Care: Tutorial. JMIR AI 2(1). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.2196/49023\u003c/span\u003e\u003cspan address=\"10.2196/49023\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWu K (2024) Optimizing Diabetes Prediction with Machine Learning: Model Comparisons and Insights. J Sci Technol 5(4):41\u0026ndash;51. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.55662/jst.2024.5403\u003c/span\u003e\u003cspan address=\"10.55662/jst.2024.5403\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003eTable .1: A summary of previous studies applying machine learning algorithms for diabetes prediction.\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eAuthors of the Study\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eUsed Algorithms\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eUsed Datasets\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eResults\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eLai \u003cem\u003eet al\u003c/em\u003e. (2019)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eLR, GBM, DT, RF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eCanadian CPCSSN, PIDD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eGBM achieved AROC 84.7%, Sensitivity 71.6%; LR AROC 84.0%, Sensitivity 73.4%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eUllah \u003cem\u003eet al\u003c/em\u003e. (2022)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eKNN, RF, XGBoost, Bagging and AdaBoost.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eBRFSS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eKNN achieved the highest accuracy: 98.38%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eKhaleel \u0026amp; Al-Bakry (2023)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eLR, NB, KNN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003ePIDD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eLR accuracy 94%, NB 79%, KNN 69%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eAlzboon et al. (2023\u003c/strong\u003e\u003cstrong\u003e\u003cspan dir=\"RTL\"\u003e(\u003c/span\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\u003cbr\u003e\n \u003cp\u003e\u0026nbsp;LR, DT, RF, KNN, NB, SVM, Gradient Boosting, Neural Network\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003ePIDD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\u003cbr\u003e\n \u003cp\u003e\u0026nbsp;Neural Network accuracy 78.57%, RF 76.30%, SVM 73.9%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eWang \u003cem\u003eet al\u003c/em\u003e. (2024)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eLASSO-regularized regression\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eAlibaba Cloud Tianchi competition\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eLASSO achieved AUC of 84.8% with ten-fold cross-validation\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eKhaledi \u003cem\u003eet al\u003c/em\u003e. (2025)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eLR, DT, RF, Bagging, AdaBoost, GBDT, SVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eShahedieh cohort\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eAdaBoost achieved the highest accuracy: 86.2%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eGhazizadeh \u003cem\u003eet al.\u003c/em\u003e (2025)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eRF, SVM, LR, DT, NB, KNN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003ePIDD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eRF highest accuracy: \u003cstrong\u003e84%\u003c/strong\u003e, SVM and LR both \u003cstrong\u003e82%\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eKiran \u003cem\u003eet al\u003c/em\u003e. (2025)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eSystematic Review\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eMultiple studies (33-year review)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eTree-based ensemble methods (Random Forest, Gradient Boosting) consistently outperformed others\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTable 2: The performance metrics of the machine learning built models for diabetes detection and prediction.\u0026nbsp;\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"628\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 102px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCategory\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eModel\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAccuracy\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eError Rate\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePrecision\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRecall\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eF-Measure\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSpecificity\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"2\" valign=\"top\" style=\"width: 102px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSimple Computational Algorithms\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003eLR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003e95.7%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e4.3%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e93.5%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e98.1%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e95.8%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003e93.2%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003eNB\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003e95.3%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e4.7%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e92.4%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e98.8%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e95.5%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003e92%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"2\" valign=\"top\" style=\"width: 102px;\"\u003e\n \u003cp id=\"_Toc196383988\"\u003e\u003cstrong\u003eTree-based Algorithms\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003eRF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003e97.5%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e2.5%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e98.7%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e96.3%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e97.5%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003e98.8%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003eGBT\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003e97.8%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e2.2%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e99.4%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e96.3%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e97.8%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003e99.4%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"2\" valign=\"top\" style=\"width: 102px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMargin-based Algorithms\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSVM\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003e96%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e4%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e94.1%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e98.1%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e96.1%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003e93.8%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003eFLM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003e93.2%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e6.8%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e90.6%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e96.3%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e93.4%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003e90.1%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"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":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Diabetes Mellitus, Machine Learning Categories, Healthcare, Clinical Prediction models, Predictive analytics","lastPublishedDoi":"10.21203/rs.3.rs-7687597/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7687597/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground:\u003c/h2\u003e\u003cp\u003eDiabetes Mellitus is a growing global health concern, projected to affect over 1.31\u0026nbsp;billion people by 2050. Early detection is vital, and machine learning offers a promising tool for predicting and managing the disease.\u003c/p\u003e\u003ch2\u003eAim:\u003c/h2\u003e\u003cp\u003eThis study aimed to introduce a structured classification of ML algorithms into three categories and to evaluate their performance in predicting diabetes using locally collected patient data.\u003c/p\u003e\u003ch2\u003eMethods:\u003c/h2\u003e\u003cp\u003eA dataset of 806 participants (403 diabetic and 403 non-diabetic) was analyzed using attributes such as sex, age, body mass index, blood glucose, blood pressure, diabetes pedigree function, and number of pregnancies (females only). ML algorithms were grouped into three categories: Simple Computational (Logistic Regression, Na\u0026iuml;ve Bayes), Tree-based (Random Forests, Gradient Boosted Trees), and Margin-based (Support Vector Machines, Fast Large Margin). Data were partitioned into training, validation, and testing sets using stratified sampling and cross-validation. Performance was assessed using accuracy, error rate, precision, recall, specificity, and F-measure.\u003c/p\u003e\u003ch2\u003eResults:\u003c/h2\u003e\u003cp\u003eTree-based algorithms outperformed other categories, with Gradient Boosted Trees achieving the highest accuracy (97.8%), followed by Random Forests (97.5%). This category also achieved superior specificity, precision, and F-measure. In contrast, Simple Computational algorithms showed the highest sensitivity (Logistic Regression 99.3%, Na\u0026iuml;ve Bayes 98.8%), effectively identifying true positive cases.\u003c/p\u003e\u003ch2\u003eConclusion:\u003c/h2\u003e\u003cp\u003eThe study\u0026rsquo;s classification framework provides a systematic basis for comparing ML models, highlighting the strengths of each category. It offers a foundation for hybrid approaches that combine high accuracy with strong sensitivity, supporting enhanced diagnostic accuracy and improved clinical decision-making.\u003c/p\u003e","manuscriptTitle":"An Evaluation of Machine Learning Categories for Diabetes Prediction and Detection in Libya: A Comparative Study","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-09-30 12:18:25","doi":"10.21203/rs.3.rs-7687597/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"0927db8f-45c0-496b-8045-1123239bdf52","owner":[],"postedDate":"September 30th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-10-05T18:53:33+00:00","versionOfRecord":[],"versionCreatedAt":"2025-09-30 12:18:25","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7687597","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7687597","identity":"rs-7687597","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}