Enhanced diabetes prediction using skip-gated recurrent unit with gradient clipping approach.

Area of Science:

• Biomedical Informatics
• Machine Learning in Healthcare
• Computational Biology

Background:

• Diabetes mellitus is a metabolic disorder characterized by hyperglycemia due to insulin deficiency or resistance.
• Machine learning (ML) offers potential for early diabetes detection, but struggles with long-term dependencies in patient data.
• Existing ML models face challenges in effectively utilizing sequential data like glucose level trends for accurate prediction.

Purpose of the Study:

• To develop a deep learning (DL) model for enhanced diabetes prediction.
• To address the limitations of conventional ML in capturing long-term dependencies in diabetes data.
• To improve the accuracy and effectiveness of early diabetes detection.

Main Methods:

• Development of a novel Skip-Gated Recurrent Unit (Skip-GRU) deep learning architecture.
• Integration of Gradient Clipping (GC) technique to stabilize Skip-GRU training and prevent exploding gradients.
• Evaluation of the Skip-GRU with GC approach on the PIMA and LMCH diabetes datasets.

Main Results:

• The Skip-GRU with GC model achieved high prediction accuracy: 98.23% on the PIMA dataset and 97.65% on the LMCH dataset.
• The proposed DL approach demonstrated superior performance compared to conventional ML methods.
• The Skip-GRU effectively captured long-term dependencies and relevant features for diabetes prediction.

Conclusions:

• The Skip-GRU with GC is a highly effective deep learning approach for accurate diabetes prediction.
• This method overcomes limitations of traditional ML in handling sequential data for disease detection.
• The approach shows significant promise for early and precise identification of diabetes mellitus.