Integrating convolutional neural networks with ensemble methods for enhanced diabetes diagnosis: a multi-dataset...

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Computational Biology

Background:

• Diabetes mellitus diagnosis faces challenges due to diverse patient data and traditional screening limitations.
• Accurate and timely diagnosis is crucial for effective diabetes management and patient outcomes.

Purpose of the Study:

• To develop a hybrid prediction framework using Convolutional Neural Networks (CNNs) and Integrated Learning with soft voting.
• To enhance the accuracy, robustness, and interpretability of diabetes diagnosis models.

Main Methods:

• A hybrid model combining CNNs for deep feature extraction with multiple classifiers (LR, SVM, RF, AdaBoost, XGBoost, LightGBM, CatBoost) via weighted soft voting.
• Evaluation on UCI Pima Indians and Tianchi Medical datasets after data preprocessing (imputation, standardization, normalization).
• Hyperparameter tuning using grid search and a 75:25 train-test split.

Main Results:

• The CNN-Voting integrated model achieved up to 98% accuracy, 0.99 F1 score, and 99% recall on the largest dataset.
• Outperformed individual models in predictive performance and generalizability.
• Identified blood glucose, BMI, age, and urea as key predictive features, aligning with clinical knowledge.

Conclusions:

• The hybrid model offers improved predictive performance and generalizability for diabetes diagnosis.
• Provides a scalable and interpretable solution for clinical decision support in diabetes management.
• Highlights the potential of integrated deep learning and ensemble methods in healthcare.