Pediatric diabetes prediction using machine learning.

Area of Science:

• Medical Informatics
• Computational Biology
• Machine Learning in Healthcare

Background:

• Diabetes is a prevalent chronic condition with significant mortality and complications.
• Limited datasets and predictive models hinder diabetes research progress.
• Big data analytics and machine learning (ML) offer solutions for diabetes prediction and classification.

Purpose of the Study:

• To develop an ML-based system for predicting diabetes likelihood and classifying its types.
• To address data limitations in diabetes research through novel dataset integration.
• To enhance early detection and management of diabetes through advanced computational methods.

Main Methods:

• Integrated four diverse datasets (paediatrics, PIMA, Pone, Gestational Diabetes) into a novel Diabetes Types Dataset.
• Employed a suite of supervised ML algorithms for multiclass diabetes type classification.
• Evaluated model performance using Accuracy, Precision, MSE, and AUC metrics.

Main Results:

• Artificial Neural Networks (ANN) achieved the highest accuracy at 99.98% for diabetes classification.
• The ML system demonstrated robust and consistent accuracy on external validation datasets.
• The system efficiently predicted 12 distinct diabetes types using a 34-feature dataset.

Conclusions:

• ML-driven approaches significantly improve diabetes detection and classification accuracy.
• The developed system supports healthcare professionals in early diabetes intervention and management.
• This research highlights the potential of big data and ML in advancing diabetes care.