Machine-Learning-Based Diabetes Mellitus Risk Prediction Using Multi-Layer Neural Network No-Prop Algorithm

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Endocrinology

Background:

• Rising global prevalence of diabetes mellitus, a chronic condition linked to high blood sugar.
• Existing diagnostic methods require enhancement for improved patient care and treatment strategies.
• Diabetes mellitus encompasses three main types: type 1, type 2, and gestational diabetes, each with distinct characteristics.

Purpose of the Study:

• To introduce a novel automated information system for classifying the three types of diabetes mellitus.
• To develop and evaluate a multi-layer neural network no-prop algorithm for accurate diabetes diagnosis.
• To enhance healthcare by providing an efficient and reliable diabetes classification tool.

Main Methods:

• Implementation of a multi-layer neural network no-prop algorithm with distinct training and testing phases.
• Utilizing an attribute-selection process to identify relevant features for classification.
• Training the neural network in a multi-layer fashion, comparing normal, type 1, type 2, and gestational diabetes.

Main Results:

• The proposed multi-layer neural network model achieved high performance metrics.
• Maximum sensitivity and specificity values of 0.97 and 0.95 were attained, respectively.
• An overall accuracy score of 97% was recorded for the classification of diabetes mellitus types.

Conclusions:

• The developed multi-layer neural network system provides a workable and efficient approach for diabetes mellitus classification.
• The system demonstrates superior performance compared to existing models, offering a promising tool for clinical application.
• This automated classification system can significantly aid in improving diabetes diagnosis and management strategies.