 In general, labeling is the process that associates and groups data that identify a specific class. In our example, the labeling task can be completed easily with the MLC tab. Data can be loaded from the GUI and labels can be specified in the text field. We are going to import all the data we just collected and specify for each group a label that identify the class. The actual feature selection is preceded by the MLC configuration. In this section, we are going to select ODR of the MLC block, which sensor should be used and what full-scale and ODRs will be set in the real application and how many device trees we decide to use. One important parameter that needs to be configured is the window length. This parameter represents the number of samples that will be used to calculate the features and execute the MLC. In our example, we will use 26 samples. This value, if we consider MLC ODR of 26 hertz, corresponds to a time window of one second. So, every second, our machine learning core will be calculating features and generate an output. Features are calculated not only from raw sensors data but can also be calculated from filters. In this section, filters can be enabled and will be available as input to the features generation block. For our example, we are going to select IPAS filter for accelerometer data and IPAS filter for gyroscope module square. Now, the features configuration can begin. With the GUI, as you can see, it's pretty easy to select the features. Select the one you think could more characterize the signal. Normally, mean, energy, variance, and peak-to-peak can be enabled by default. Do not worry to enable too many features. Training process will get rid of the ones that they are considered useless. For our example, we are going to select mean, variance, energy, and peak-to-peak. This concludes the feature selection. We are now able to generate the ARFF file that can be used by WACA, for instance, for the training process.