JUIndoorLoc: A Ubiquitous Framework for Smartphone-Based Indoor Localization Subject to Context and Device Heterogeneity

Article Type

Research Article

Publication Title

Wireless Personal Communications


A new era of ubiquitous indoor location awareness is on the horizon especially for context sensing, ambient assisted living and many other smart city applications. Although indoor localization plays a pivotal role in making the environment smarter, it is still very difficult to compare state-of-the-art localization algorithms due to the scarcity of standard databases. Publicly available databases are neither fine-grained nor contain data for different conditions. Received Signal Strength Indicator (RSSI) of Wi-Fi signals vary with indoor environment (open/closed room, presence/absence of user, temperature etc.) and scanning smart hand-held devices. Thus, localization accuracy varies with various environmental conditions and also granularity of location points (cell). Consequently, in this paper, our contribution is two-fold. First, we present a comprehensive indoor localization dataset, subject to different domains-spatial, temporal, context and device. RSSI data has been collected with cell sizes as small as 1m×1m from three floors of a building of our University using an Android application built for this purpose. This multi-floor dataset is available online at https://drive.google.com/open?id=1_z1qhoRIcpineP9AHkfVGCfB2Fd_e-fD. Our experimental results show that maximum of 71.78 % classification accuracy can be achieved for state-of-the-art classifiers when training and testing samples are taken in different environmental conditions and from smartphones having different configurations. Single classifier cannot easily be modified to suit these variations without loosing its generality. So, to overcome these conditional dependencies, our second contribution is to propose a framework for indoor localization, JUIndoorLoc and design an ensemble of condition specific classifiers as part of the framework to take care of context and device heterogeneity. Consequently, this ensemble of condition specific classifiers is implemented and found to predict a location with 91.74 % accuracy (1.87 m) for our dataset.

First Page


Last Page




Publication Date