Automated systems that can accurately diagnose diabetic

The gap

Evidence profile

Sourced from the limitations section and future work and recommendations and conclusions and stated research gap of the source papers, classified as general, spanning 5 journals.

Supporting evidence — 6 representative gaps

• Early Type 2 diabetes risk prediction using explainable machine learning in a two-stage approach (2026) · Frontiers in Digital Health · doi

  The models developed in stage one were trained on a relatively small dataset, which may increase the risk of overfitting. The pronounced class imbalance in stage two required the application of SMOTE, which may have introduced bias into the training data. The relatively limited number of features available in stage two may have constrained the models' ability to fully capture the underlying predictors of diabetes risk.

  generallimitations section
  Keywords: models developed stage one were trained relatively small
• Early Type 2 diabetes risk prediction using explainable machine learning in a two-stage approach (2026) · Frontiers in Digital Health · doi

  This study employed a two-stage approach to train and evaluate five machine learning models for predicting diabetes risk. The MLP demonstrated superior performance in predicting type 2 diabetes risk in stage one with an accuracy of 95.19%. The model for this stage is designed to predict type 2 diabetes risk based on symptoms. Using SHAP methods, the top ten most influential features were identified. Retraining all models with only these features yielded an accuracy of 95.19%, which is the same as that achieved by the MLP model when trained on all features, indicating the effectiveness of the feature selection methods employed. In stage two, the XGBoost model achieved an accuracy of 97.14% in identifying referable cases of diabetes using data collected using the mUzima application. SHAP analysis provided global and local model explanations, ranking diastolic blood pressure as the most critical predictor, followed by waist circumference. Based on recommend integrating the models of both stages into the mUzima app to support community health workers in efficiently identifying high-risk individuals. Furthermore, we encourage the use of comprehensive data recording to enhance model performance through richer feature sets, including socio-economic features. these findings, we Future work will focus on deploying stage one and stage two into the mUzima mobile application and integrating them with OpenMRS to support the identification of referable diabetes cases. Additionally, incorporating diabetes diagnoses using explainable machine learning into mUzima by health centers will enable continuous model updates and improvement.

  generalfuture workevidence 5/5
  Keywords: stage diabetes model risk using features muzima models accuracy employed machine learning predicting performance type
• Feature-Optimized and Explainable Machine Learning Framework for Early Diabetes Prediction Using Hybrid Clinical and Lifestyle Data (2026) · International Journal of Scientific Engineering and Research · doi

  • learning algorithms • Employs several machine in analyzing and comparing various models for the purpose of choosing the most accurate algorithm for predicting diabetes. Implements methods of feature selection which ensure high-quality data and consequently increase the accuracy of the model through the use of significant variables. • Aids in determining significant health factors such as blood sugar levels, body mass index, and age, which affect diabetes prediction. • Assists in early prediction of diabetes risks for purposes of seeking early medical intervention. • The entire process can be regarded as relatively simple and efficient.

  generalfuture workevidence 5/5
  Keywords: diabetes significant prediction early learning algorithms employs several machine analyzing comparing various models purpose choosing
• Predictive modeling of early diabetes diagnosis: An evaluation of XGBoost, support vector machine, and random forest classifiers (2026) · International Journal of Science and Research Archive · doi

  To further strengthen the predictive accuracy of diabetes diagnosis models, it is advisable to incorporate a wider array of clinical and lifestyle features. Adding biomarkers such as blood glucose levels, HbA1c, and insulin measurements could provide more nuanced information about an individual's metabolic health. Moreover, including longitudinal data that tracks changes over time may allow the model to detect early trends in diabetes progression, facilitating timely intervention before the condition advances. Future research should also prioritize optimizing model performance through advanced strategies such as hyperparameter tuning, systematic feature selection, and ensemble methods. The use of automated machine learning (AutoML) frameworks could simplify this process by quickly identifying optimal model configurations. Additionally, 876 International Journal of Science and Research Archive, 2026, 20(01), 865–878 exploring deep learning techniques may offer further enhancements, particularly with larger, more complex datasets, as these architectures are adept at capturing intricate nonlinear relationships among variables. Another crucial recommendation involves effectively addressing class imbalance, as current models tend to be less sensitive to diabetes-positive cases. Methods like synthetic minority oversampling (SMOTE), cost-sensitive learning, or tailored loss functions can help rebalance the training dataset and improve detection of minority classes. Reducing false negatives is particularly vital in medical contexts, where failing to identify positive cases can have serious health implications. Finally, for real-world implementation, these models should be integrated into user-friendly platforms and connected with existing electronic health record systems. Engaging healthcare professionals in the design process will help create practical, actionable decision-support tools. Continuous validation using updated patient data, coupled with feedback from clinical deployment, will ensure that the models remain accurate, relevant, and valuable for early diabetes detection over time.

  generalrecommendationsevidence 5/5
  Keywords: diabetes models health model learning further clinical time early process particularly sensitive positive cases minority
• An XGBoost-based predictive framework for diabetes mellitus multi-classification (2026) · Scientific Reports · doi

  Future research may focus on validating the proposed model using larger clinical datasets to extend the generalization capability of the proposed model. Additionally, hybrid imbalance-handling could be investigated to further improve minority subtype detection, particularly in multiclass scenarios where certain diabetes subtypes are underrepresented.

  generalconclusionsevidence 5/5
  Keywords: proposed model future focus validating using larger clinical datasets extend generalization capability additionally hybrid imbalance
• Deep Learning aplicado ao diagnóstico assistido de retinopatia diabética (2026) · Revista ft · doi

  There is a need for automated systems that can accurately diagnose diabetic retinopathy. Prior work has focused on developing automated systems, but there is still a need for more accurate and reliable models. The use of deep learning has emerged as a promising approach, but more research is needed to fully explore its potential.

  generalstated research gapevidence 4/5
  Keywords: there need automated systems accurately diagnose diabetic retinopathy

Questions about this gap