Multi-Sensor Data Fusion for Occupancy Detection Using Dempster–Shafer Theory

Abstract

A data fusion technique has been proposed in this paper for detecting the presence and absence of individuals in a room. It involves the utilization of data from a series of sensors mainly temperature and humidity sensors, for detecting the presence of a person. From the perspective of evidence theory, data collected from every sensor can be viewed as a matter of evidence. A Dempster–Shafer (D-S) theory-based data fusion model is established for modeling and consolidating the pieces of evidence and hence generating an overall speculation of the temperature and humidity level of a room. Testing has been carried out with a dataset that have two classes. At first, the detection is done using various well-known classifiers such as logistic regression which shows an accuracy level of 94%, K-nearest neighbors shows an accuracy of 93%, support vector machines result in an accuracy of 94%, and decision tree classifier and random forest classifier show an accuracy of 92% and 93%, respectively. A subset of the data is used to create class membership probabilities for every attribute for training, and hence, a mass function is created. Finally, D-S rule is applied, and the outcomes suggest that the data fusion method gives a higher accuracy level compared to the only involvement of classifiers.