Innovative Deep Learning Strategies for Chaotic Data Classification: A Multi-Algorithm Comparison in the Presence of Noise

Abstract

Chaos is prevalent in both nature and science, appearing in data, time series and complex systems. Chaotic systems exhibit numerous uncertainties, akin to noise, which challenge researchers to distinguish or analyze potential underlying patterns or even identify the type of system involved. However, determining the kind of chaotic system is essential, as it enables prediction, synchronization, control, treatment and application. This study employs machine learning to classify chaotic data through a simulation involving three types of research data: Lorenz data, Lorenz combined with Gaussian white noise, Gaussian white noise and pink noise, utilizing six distinct algorithms. The most effective testing results are achieved using Mobilenet, with a classification accuracy of 97.38% and a loss of 0.2680 across these six data types.