Classification of Chaotic Dynamics Through Time–Frequency Representations and Machine Learning

This study employs an approach that centers on the utilization of support vector machine classifiers to categorize chaotic dynamics. Commonly, within chaotic secure communication schemes, one state variable of a determined dynamical system is employed for encrypting information, while another operates as the cryptographic key. Successful communication requires the receiver to synchronize using this key state. In the event of a security breach or unauthorized access to this data, the ability to accurately identify the specific chaotic system and its state becomes essential for preventing potential attacks. This chapter presents an approach to the classification of distinct chaotic dynamics based on time-domain data observation. The proposed method combines Time–Frequency Representations (TFR) with machine learning, utilizing Support Vector Machine (SVM) classifiers, resulting in effective classification of both the chaotic system and its state. This approach extends to multiclass classification, where we employ an error-correcting output codes approach to decompose complex tasks into simpler, binary problems, thereby enhancing both scalability and versatility, as indicated by the results demonstrating good performance.