Abstract
WiFi fingerprinting indoor positioning systems have extensive applied prospects. However, a vast amount of data in a particular environment has to be gathered to establish a fingerprinting database. Deficiencies of these systems are the lack of universality of multipath effects and a burden of heavy workload on fingerprint storage. Thus, this paper presents a novel Random Forest fingerprinting localization (RFFP) method using channel state information (CSI), which utilizes the Random Forest model trained in the offline stage as fingerprints in order to economize memory space and possess a good anti-multipath characteristic. Furthermore, a series of specific experiments are conducted in a microwave anechoic chamber and an office to detail the localization performance of RFFP with different wireless channel circumstances, system parameters, algorithms, and input datasets. In addition, compared with other algorithms including K-Nearest-Neighbor (KNN), Weighted K-Nearest-Neighbor (WKNN), REPTree, CART, and J48, the RFFP method provides far greater classification accuracy as well as lower mean location error. The proposed method offers outstanding comprehensive performance including accuracy, robustness, low workload, and better anti-multipath-fading.
Highlights
With the increasing proliferation rate of smart devices, Location Based Services (LBSs) have gained considerable attention
We tackled the problem of channel state information (CSI)-based localization system by the Random Forest fingerprinting localization (RFFP), which is a Random Forest-based indoor fingerprinting system that utilizes Feedback Decision Trees (FDT)
A series of specific experiments were conducted in a microwave anechoic chamber and an office to detail the localization performance of RFFP with different system parameters, algorithms, and input datasets
Summary
With the increasing proliferation rate of smart devices, Location Based Services (LBSs) have gained considerable attention. A subset is randomly extraced from all attributes of CSI values as the training data set to construct a fast decision tree; the size of the RF determines the number of DTs. Second, a root mean squared error (RMSE)-based pruning strategy is applied to reduce the computational complexity as well as overcome the overfitting problem and improve the precision rate of DTs. Third, the votes in the output of every decision tree are counted, and the most frequently occurring class number is the final classification result of the proposed method. The improved Random Forest-based fingerprinting localization method possesses the large number of virtues mentioned previously and discovers the feature of wireless channel data.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
