A region-wise indoor localization system based on unsupervised learning and ant colony optimization technique

Abstract

WiFi-based indoor localization has gained widespread attention in the recent past with the ubiquitous deployment of WLAN. However, for sustainable performance, it is crucial to identify and maintain the important WiFi Access Points (APs). Interestingly, the localization capabilities of APs differ even within their area of coverage depending on the indoor ambience and building properties. Thus, the significance of an AP to the localization performance can be better assessed if the entire region is divided into optimal number of clusters/subregions. This type of problem is hardly investigated in the literature. Consequently, in this paper, the aforementioned challenges are addressed from the perspective of expert systems through applying machine learning and meta-heuristic techniques. Accordingly, our contribution is three-fold. First, we have designed a Region-wise Indoor Localization System (RwILS) in which a sub-regional division algorithm is proposed using DBSCAN approach. Second, a sub-region-wise important AP selection algorithm is designed based on Ant Colony Optimization technique. Third, a location estimation approach is proposed that first identifies the sub-region and then predicts the location point in that sub-region using supervised learning classifiers. A publicly available dataset, JUIndoorLoc is used for experimental evaluation. The localization accuracy of RwILS is improved to 95.68% while the state-of-the-art classifiers give 71% to 83% accuracy. RwILS also outperforms some recent works which further validates its effectiveness. Thus, this proposed technique can lead to an effective expert system in the domain of indoor localization.