Room-Level Activity Classification from Contextual Electricity Usage Data in a Residential Home

Human activity recognition is challenging without compromising users' privacy and burdening them with wearable devices, cameras, mobile applications, etc. As the smart energy meter usage trend is increasing worldwide, it can be used as a non-invasive activity monitoring methodology without violating users' privacy and requiring an additional installation cost where smart energy meters are already in use. In addition, household energy consumption patterns, including the consumed power, current intensity, and energy usage, are mainly determined by the individual's needs, lifestyle, and time context, which can offer important information about the household's daily activities. Using energy data, users can get information about ongoing activities in each room of the house under observation. This paper uses different machine-learning (ML) algorithms such as Random Forest, Decision Tree, K-Nearest Neighbour, and Support Vector Machines for activity recognition from load classification. The ML model classifies different real-time activities in the same room based on the consumed power estimates. By utilizing an open smart energy sub-meter dataset, activity patterns of household occupants are identified. The load classification analyses, employing the aforementioned ML algorithms, demonstrate an activity recognition accuracy of up to 99%.