Indoor localization framework with WiFi fingerprinting

Indoor localization through WiFi fingerprinting requires a large number of fine-grained data samples. This study presents a data acquisition and indoor localization framework that collects crowd-sourced WiFi received signal strength data in a metropolitan high-rise building and predicts location through WiFi fingerprinting. The framework consists of a server and an Android application and was tested at NYIT for data collection for two weeks in December 2016. The dataset was preprocessed and analyzed through linear support vector machine to test location prediction accuracy. Various feature selection schemes were compared for their location prediction accuracy. We show that a small subset of features suffices to provide high location prediction accuracy. The average location prediction accuracy increases from 83% to 100% when time features are considered comparing to using only spatial features.