Pediatric diabetes prediction using deep learning.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Computational Biology

Background:

• Early diabetes prediction is crucial for effective management and prevention of complications.
• Deep Learning (DL) models have shown promise in accelerating diabetes diagnosis.
• Existing methods require optimization for improved accuracy and efficiency.

Purpose of the Study:

• To propose a novel technique for highly accurate early diabetes prediction using Deep Learning.
• To develop and validate a Deep Neural Network (DNN) model for diabetes diagnosis.
• To optimize the DL model's performance through meticulous hyperparameter tuning.

Main Methods:

• Implementation of a Deep Neural Network (DNN) with ten hidden layers and multiple epochs.
• Utilizing the Deep Neural Network (DNN)-based multi-layer perceptron (MLP) algorithm.
• Fine-tuning hyperparameters for automated data preprocessing, prediction, and classification on the MUCHD dataset.
• Validation using cross-validation with metrics including accuracy, F-score, precision, sensitivity, specificity, and Dice similarity coefficient on 548 patients with 18 features.

Main Results:

• The proposed DL system achieved a remarkable accuracy rate of 99.8% for diabetes prediction.
• The model demonstrated a 0.39% increase in overall system performance compared to state-of-the-art methods.
• Rigorous validation using multiple metrics confirmed the reliability and high performance of the prediction system.

Conclusions:

• The developed DL technique offers a highly accurate and reliable method for early diabetes prediction.
• The proposed DNN-based MLP model shows significant potential for clinical application in diabetes diagnosis.
• This novel approach represents an advancement in leveraging AI for improved diabetes care and management.