ADL recognition through machine learning algorithms on IoT air quality sensor dataset

The Human Activity Recognition is a focal point for Ambient Assisted Living, and may be implemented in several ways usually involving the use of different technologies, as wearable, video, environmental or radio frequency sensors, which can be used alone or in combination among them. Recently, the approaches based on machine learning have attracted a lot of interest, especially in order to create recognition systems that do not require a high detection capacity by the single sensor, as they base their decision on the processing of the information acquired from multiple sensors simultaneously. The aim of the present work is to derive information about the activities that are carried out inside the house on the basis of the data acquired by a set of sensors analyzing the air components. The Human Activity Recognition is then the result of a machine learning classification of the output of an array of low cost “commercial off-the-shelf” air quality sensors. The considered recognition system exploits electrochemical sensing, Wi-Fi technology, cloud computing, machine learning and application services. The obtained results evidence that a good accuracy in the recognition of “activities of daily living” is reached, even if a not calibrated sensing was performed.