Epidemic dynamics prediction using fractional SIRD and deep learning

This paper introduces, analyzes, and numerically investigates a fractional-order SIRD epidemic model employing the normalized Caputo-Fabrizio (NCF) derivative. The model incorporates memory effects and explicitly tracks disease-induced mortality through a deceased compartment, offering a more comprehensive framework for epidemic dynamics. We establish existence, uniqueness, positivity, and population conservation for the proposed system, and present a robust numerical scheme. We demonstrate the influence of the memory parameter and kernel normalization through simulations, and highlight their significance for epidemic forecasting and real-world applications. Deep neural network (DNN) was used to approximate SIRD dynamics with fractional order and improve the accuracy of predictions. The model successfully extracted nonlinear time-dependent patterns from simulated epidemic data through its learning process. The proposed framework demonstrated strong predictive ability, achieving mean square error values as low as 0.00027 and root mean square errors below 0.17 across different compartments. By integrating fractional calculus with deep learning techniques, the model establishes a reliable approach for representing complex disease dynamics and provides meaningful insights that can support public health decision-making.