Modeling chaotic diabetes systems using fully recurrent neural networks enhanced by fractional-order learning

Area of Science:

• Biomedical Engineering
• Computational Neuroscience
• Control Systems

Background:

• Nonlinear medical systems, like those governing diabetes, exhibit complex chaotic dynamics.
• Artificial Neural Networks (ANNs) are effective tools for modeling such intricate biological systems.
• Accurate modeling is crucial for understanding and managing diseases like diabetes.

Purpose of the Study:

• To develop an advanced artificial neural network model for accurate simulation of chaotic diabetes systems.
• To enhance the modeling capabilities of Fully Recurrent Neural Networks (FRNNs) using Fractional-Order (FO) learning.
• To evaluate the proposed model's performance against existing methods in simulating the insulin-glucose regulatory system.

Main Methods:

• Utilized a Fully Recurrent Neural Network (FRNN) integrated with a Fractional-Order (FO) learning algorithm.
• Implemented a Lyapunov-based mechanism for deriving online learning rates to ensure stability and adaptive tuning.
• Simulated the insulin-glucose regulatory system across various diabetic conditions (Type 1, Type 2, hyperinsulinemia, hypoglycemia).

Main Results:

• The proposed FRNN-FO model achieved higher accuracy and robustness compared to traditional FRNN-GD, DFNN, DRNN-GD, and DRNN-FO models.
• Fractional-order learning significantly improved the network's modeling accuracy and convergence speed.
• The Lyapunov-based adaptive learning mechanism ensured stable and efficient parameter tuning.

Conclusions:

• The FRNN-FO model presents a powerful and accurate approach for modeling complex nonlinear biomedical dynamics, particularly in diabetes.
• This advanced modeling technique offers a promising tool for clinical research and personalized diabetes management.
• The integration of fractional-order calculus and neural networks opens new avenues for understanding and treating complex diseases.