Virtual Sensors Method and Architecture for a Smart Home Environment with Random Forest

A smart home environment can be built with several functions at once. However, the conventional integration of several systems involves many devices and sensors, resulting in expensive and complex solutions. Our research aim is to evaluate the performance of a simplified solution using sensor virtualization on smart homes with random forests. We use the smart home dataset from Kaggle, which has carbon monoxide (CO) data, humidity, light status, liquefied petroleum gas (LPG) content, motion status, smoke, and temperature. First, we designed a smart home with an air quality measurement system, smart lighting, and a fire detector. Then we design a virtual sensor that can measure CO, light status, LPG content, and smoke from humidity, temperature, and motion status using a random forest. We use linear regression and adaptive booster (AdaBoost) as benchmark methods. Our test metrics are root mean squared error (RMSE) and
$r^{2}$
. The test results show that the random forest method's virtual sensor performs better than linear regression and AdaBoost with RMSE and
$r^{2}$
values of 0.029 and 0.803, respectively.