“…In this regard, the fast Fourier transform [ 17 , 18 ], statistical methods [ 19 , 20 ], Welch method [ 21 ], regressive-based models [ 22 ], fractality-based method [ 23 ], entropy-based methods [ 24 , 25 ], multiple signal classification method [ 26 ], wavelet transform [ 27 , 28 , 29 ], empirical mode decomposition [ 30 , 31 ], and principal component analysis [ 32 ], among other indices or methods, have been explored to extract patterns about the IM condition. In a similar venue, different pattern recognition algorithms have already been presented to diagnose the IM condition automatically, e.g., artificial neural networks [ 4 ], fuzzy logic systems [ 23 ], k-means [ 33 ], support vector machines [ 34 ], and decision trees [ 35 ], among others. Notwithstanding the obtaining of promising results in the above-mentioned works, those techniques or algorithms present diverse issues that can compromise their performance in real-life situations, for instance: (1) a fine-tuning (a procedure performed typically by trial-and-error) of diverse parameters such as decomposition level, wavelet mother, model order, among others, for properly analyzing the in-test signals is required [ 36 ]; (2) noisy signals with nonstationary properties as the ones measured in the IMs degrades somehow their performance [ 37 ]; and (3) the adroit integration of feature (or set of features) and classifier is achieved by trial and error, where in all the cases the researcher proposes, tests, and selects the features to be used, which, on the one hand, increases the complexity and, on the other hand, might not lead to the best results [ 15 ].…”