Abstract
Coffee plants are susceptible to a variety of diseases that threaten the quantity and quality of coffee
production. Coffee Berry Disease (CBD) poses a significant threat to global coffee production,
particularly in Africa, and creates a potential risk of expansion into coffee-growing regions in Latin
America and Asia. This study explores a novel hybrid computational modeling using fractional
calculus and deep learning techniques to model the complex dynamics of CBD. The Caputo
operator-based Euler scheme is used to solve the model numerically and to generate synthetic
datasets by considering various values of fractional order. Further, the dynamics of disease
transmission are numerically studied using a deep neural network by comparing solutions across all
compartments through convergence testing, regression metrics and error distribution. To minimize
the mean squared error, we employ a fractional Euler iterative scheme, allocating 70 % of the data
for training and 15% each for validation and testing. Numerical performance is demonstrated under
five fractional-order scenarios (α = 0.80,0.85,0.90,0.95,1) using distinct initial conditions. The
results obtain from present approach align precisely with the benchmark data. The best validation
performance is archived to 10−10, and with minimum absolute error 10−8. The outcomes of the
scheme are rigorously presented numerically and tabulated. We believe that the integration of a
deep neural network that implements Tanh and ReLU activations in its hidden layers to study CBD
is a novel attempt in the modeling of infectious diseases. This hybrid neuro-computing paradigm
delivers both improved accuracy and computational efficiency, providing a better understanding of
CBD and thereby helping to set effective control interventions.
Keywords: Coffee berry disease modeling; Fractional calculus; Deep learning approach; Error
estimation; Memory-based simulation.