Font Size: 

Many physical systems doesn’t reveal significant disruptions when subjected to small parametric variations, which may represent a challenge to assess these values. In such cases, it is crucial to employ sensitive methods to assess parameter values based on the system’s observed data. Recurrence analysis becomes explicit in many data properties, such as being chaotic or random. This analysis can be performed on linear and nonlinear systems. The information obtained by using recurrence analysis can be used to detect changes, classify, and even forecast the dynamic evolution. We propose an approach that integrates recurrence analysis with supervised machine learning techniques. This method involves utilizing recurrence quantifiers: microstate probabilities, entropy, and the threshold, associated with a multi-layer perceptron to evaluate parameters governing the dynamics of systems. We show the effectiveness of the method in determining parameters of chaotic maps, using the generalized Bernoulli, the so-called β-x, mod 1, logistic, and Hénon maps. The MMLP distinguishes minor parameter changes and our findings indicate that, for machine learning classification, employing higher values of the recurrence plot threshold ε associated with the system’s average maximum entropy, yields superior results compared to smaller thresholds commonly recommended in the literature.