RSSI -based Classification for Indoor Localization in Wireless Sensor Networks

Indoor localization is an important issue for a very large number of applications. Localization systems based on received signal strength indicators (RSSI) have become very popular in recent years due to their simple and low cost implementation. For most applications, contextual information related to the target's region is sufficient. Therefore, in this paper, instead of estimating the exact coordinates of the target, we propose an approach to determine its region by considering the localization as a multi-class classification problem. State-of-the-art multi-class classification techniques are investigated such as the K-nearest neighbor (KNN) algorithm, one-vs-all logistic regression and one-vs-all support vector machines (SVM). To compare the algorithms and evaluate their performance, we conduct our experiments on real RSSI datasets collected from a real environment implemented using a Zigbee network.