Quality of data logs is a key point for Machine Learning based algorithms, since the model is built from the data.
It is important to correctly acquire and label data. For instance, if the class "walking" has to be recognized, the data log related to this class should contain only real walking data, In other words, the acquisition should start when the user is already walking, and the user should not stop walking at all during data logging. In some cases data cleaning might be necessary.
