iGRM

Abstract

Occupancy detection is one of the many applications of Building Automation Systems (BAS) or Heating, Ventilation, and Air Conditioning (HVAC) control systems, especially, with the rising demand of Internet of Things (IoT) services. This article describes the fusion of data collected from sensors by exploiting their potential to sense occupancy in a room. For this purpose, a sensor test bed is deployed that includes four sensors measuring temperature, relative humidity, distance from the first obstacle, and light along with a Arduino micro-controller to validate our model. In addition, this article proposes three algorithms for efficient fusion of the sensor data that is inspired by the Grey theory. An improved Grey Relational Model (iGRM) is proposed, which acts as the base classifier for the other two algorithms, namely, Grey Relational Model with Bagging (iGRM-BG) and Grey Relational Model with Boosting (iGRM-BT). Furthermore, all three algorithms use a sliding window concept, where only the samples inside the window participate in model training. Also, we have considered varying number of window size for optimal comparison. The algorithms were tested against the experimental data collected through a test bed as well as on a publicly available large dataset, where both the ensemble models, iGRM-BG and iGRM-BT, are seen to enhance the performance of iGRM. The results reveal exceptionally high performances with accuracies above 95% (iGRM) and up to 100% (iGRM-BT) for the experimental dataset and above 98.24% (iGRM) and up to 99.49% (iGRM-BG) using the publicly available dataset. Among the three proposed models, iGRM-BG was observed to outperform both iGRM and iGRM-BT owing to its advantage of being an ensemble model and its robustness against over-fitting.