Impact of neuron count and learning rate on the accuracy of a NumPy-based artificial neural network in predictive modeling of nonlinear dynamical systems
Authors: David Mark M. Mueller
Volume 1, No. 1 •
Published: 2024-12-04 •
Open Access
Abstract
The impact of neuron count and learning rate on the accuracy of an artificial neural network (ANN) in predictive modeling of nonlinear dynamical systems is explored. This study is focused on the Lorenz, Rössler, and Chen systems, which are renowned for their sensitivity to initial conditions, intricate dynamics, and strangely attractive plot of their trajectories. The model's performance was assessed using symmetric mean absolute percentage error (SMAPE) and coefficient of determination (R²). The results shows that the model with 24 neurons and 0.1 learning rate consistently outperformed other parameters across all three systems.
Keywords
chaotic system, forecasting, machine learning, complex dynamics