Smartwatch activity recognition feature comparison using ML.net

Smart environments are more accessible than ever and are created by a synergy of different devices that provide smart capabilities and can work together. Activity recognition is an important part of the smart environment, as it allows the system to know what activities are the occupants doing. One of the simplest options to do activity recognition is to use body-worn sensors, and the smartwatch, in this case, proves to be a good candidate. Our objective is to compare the precision of the activity recognition process using accelerometer or gyroscope extracted features data. For this, we use the ML.net machine learning framework to analyze 18 different activities.