Creating Refined Datasets for Better Chaos Detection.

Area of Science:

• Signal analysis
• Dynamical systems
• Chaos theory

Background:

• Analysis of signal properties, particularly biomedical signals, is a growing research area.
• Identifying chaotic properties within signals is a key challenge in signal analysis.
• Existing methods rely on synthetic signals from known systems to train chaos detection classifiers.

Purpose of the Study:

• To propose a new method for generating and extracting signals to improve chaos detection classifiers.
• To create and publicly release referential datasets of refined signals for training and validation.
• To leverage the sensitive dependence on initial conditions characteristic of chaotic systems.

Main Methods:

• Reconstructing multidimensional phase space from signal data.
• Applying data clustering techniques to group signals with similar initial conditions.
• Generating groups of signals with subtle variations in initial conditions to highlight chaotic behavior.

Main Results:

• A novel method for obtaining/extracting signals that effectively trains classifiers to detect chaos.
• Creation of publicly available, referential datasets of refined signals.
• Demonstrated usefulness of the new datasets through an experiment using a Long Short-Term Memory (LSTM) neural network.

Conclusions:

• The proposed method successfully generates signals that enhance chaos detection capabilities.
• The new datasets provide valuable resources for advancing research in signal analysis and chaos detection.
• The approach is effective in training machine learning models, such as LSTM networks, for identifying chaotic dynamics.