An Empirical Evaluation of Machine Learning Algorithms for Indoor Localization using Dual-Band WiFi

WiFi based indoor localization systems relying on received signal strength indicator (RSSI) have attained large acceptance over the past few years as mobile devices with WiFi capability are prevalent in everyday routines and practices, this is in view of the demand for low-cost indoor positioning systems (IPSs). Extant RSSI reliant WiFi based IPSs utilize raw RSSI of signals received from access points (APs) to evaluate device locations. However, the particular raw RSSI of signals may readily fluctuate and may be susceptible to interference as a consequence to multipath propagation channels, other wireless local area networks, device diverseness, and noise. To overcome these prevailing issues, we investigate performance enhancement of WiFi fingerprinting-based IPSs for increased accuracy and robustness in positioning over dual-band WiFi (such as IEEE 802.11n) that employs both of the 2.4GHz and 5GHz frequency spectrum bands as a conceivable substitute. To that end, based on empirical concurrent measurements, we conduct a comparative performance analysis of a collection of machine learning (ML) classification algorithms to evaluate their classification capacities in determining the location of a WiFi receiver device in a single and a dual frequency band operation setting. Numerical results demonstrated that in our IPS, the location could be effectively predicted by means of a subset of the collection of considered ML classification algorithms when using the raw RSSI of both of the 2.4GHz and 5GHz frequency bands. Computed evaluation metrics to characterize performance of the IPS served to identify an optimum ML algorithm based on attained results to accurately localize the device, and the existence of dual frequency information renders the positioning process to be more robust.