Human gait is not a periodic movement, its analysis will always present a typical intrasubject, intersubject variability. For this reason recent study enlighted the importance of taking in account a large number of gait cycles to be analyzed, making easier to compare parameters. From this point of view the detecting of gait deviations can be more accurate than visual assessment and can be used in different clinical cases. Although the existence of many protocols for recording gait signals, the selection of signal's features is the main issue when a machine learning algorithm is used to fit a designed model. In this study an organized data set composed by gait parameters observations, is processed through unsupervised and supervised learning algorithms in order to predict an asymmetry level assigned to each observation through the EMG asymmetry index proposed in a recent study.