Hyperparameters optimisation for time varying signals

For the most machine learning methods, for cyclo-stationary or even stochastic signals, the performance depends critically on hyperparameters. Moreover, the tuning of more hyperparameters based on the feedback of the performance model will leak an increasingly significant amount of information about the validation set into the model. Therefore, we propose in this research two classes of hyperparameters, a general class that makes the characterization of general signal curve and the second, a specific class that define special parameters connected to the phenomena type (e.g. sensor type).