Dr. Zulqurnain Sabir is an Assistant Professor of Mathematics at the Lebanese American University, Beirut, Lebanon, with extensive experience in teaching, research, and academic development. He earned his Ph.D. in Mathematics from Hazara University in 2022, following his M.Phil and M.Sc. degrees in Mathematics, and has further strengthened his research profile through postdoctoral positions at United Arab Emirates University and Harran University, Turkey. Dr. Sabir has a diverse academic and professional background, having served as a lecturer at the University of the Punjab (Jhelum Campus) and as a Senior Researcher at United Arab Emirates University. He also demonstrated leadership in education by founding and directing institutions focused on mathematics and science education. He has actively supervised Ph.D., M.Phil, and undergraduate students, contributing to numerous research publications. His research focuses on mathematical modeling, fractional and differential systems, and numerical computing, with applications in biological and dynamical systems. Recognized for his contributions to science and research, Dr. Sabir received the International Scientist Award in Engineering, Science, and Medicine in 2021. His expertise is supported by strong computational skills, including MATLAB, Mathematica, LaTeX, and advanced numerical methods.

A Novel Three-Layered Neural Network with Mixed Activation Functions for Simulating Prey-Predator Dynamics The aim of current work is to provide the numerical solutions of the predator mathematical system (PPMS) by designing a three layered neural network (TLNN) procedure. The PPMS is one of the dynamic models that simulates the interactions between two species: predator and prey. The designed neural network structure contains three hidden layers with 10, 20, and 30 numbers of neurons, operating a log-sigmoid function in the first hidden layer, poitive linear transfer function in the second hidden layer, and tan-sigmoid activation function in the third hidden layer to capture the nonlinear dynamics. The training is performed by the Broyden-Fletcher-Goldfarb-Shanno (BFGS) quasi-Newton scheme, which is one of the popular optimization methods used in neural networks to upgrade the biases and weights. An explicit Runge-Kutta is used to get the dataset, which is trained furhter to train through BFGS quasi-Newton scheme. The data is divided into training, validation, and testing with reasonable percentages in order to reduce mean square error. The performance of the scheme is evaluated through overlapping of the calculated and reference outcomes, demonstrating its accuracy. The small absolute error and perfect correlation coefficient (R = 1) further confirm the model’s reliability. Additionally, transitions of state, histogram error, and best values of the training data validate the effectiveness of the proposed scheme for solving the PPMMS Keywords: Three-layered neural network; Predator-prey; Log-sigmoid; BFGS quasi-Newton scheme; Tan-sigmoid.