PEWFIPS-HV

Evaluating Wi-Fi Fingerprinting for Enhanced Indoor Positioning in Campus Environments

Wi-Fi fingerprinting indoor positioning systems (FPIPS) are vital for delivering location-based services in indoor environments where GPS is impractical. However, their effectiveness relies on selecting appropriate algorithms. This study stresses the importance of evaluating Wi-Fi FP-IPS algorithms, especially in campus environments where precise localization is crucial. We conducted quantitative experiments to assess the performance of these algorithms at University West’s indoor campus. Our analysis, including data collection and implementation of knearest neighbors (KNN) algorithms, focused on key metrics like accuracy, precision, and computational cost. Our findings enhance indoor positioning accuracy and provide valuable insights for engineers implementing Wi-Fi FP-IPS, highlighting the superiority of the region-based algorithm for real-world applications.

INTRODUCTION

GPS is widely used for positioning and navigation, offering accuracy within approximately 5 meters. However, its performance is impaired in environments obstructed by nonline of sight (NLOS) barriers like buildings, limiting its effectiveness indoors. To overcome this, indoor positioning systems (IPS) have been developed, utilizing technologies such as magnetic, infrared, ultrasonic, ultrawide band, Bluetooth, and Wi-Fi, with Wi-Fi emerging as a popular choice due to its existing infrastructure. Fingerprinting is commonly employed in Wi-Fi-based IPS, where users capture data signals emitted by access points (APs), and an algorithm compares them against a fingerprinting database to estimate the user’s position. Various algorithms are applicable in developing a Wi-Fi IPS, with notable ones including K-nearest neighbor (KNN), random forest, artificial neural network (ANN), support vector machine (SVM), and others. Each algorithm has distinct advantages and drawbacks, particularly in terms of accuracy and computational overhead. KNN operates by selecting the K-nearest reference points (RPs) and calculating the Euclidean distance to determine the closest RP. Among these methods, KNN-based algorithms are favored due to their simplicity and scalability. Conversely, AI and ML algorithms offer heightened accuracy but require more computational resources and training periods. However, KNN algorithms can be readily implemented with existing RP datasets. Therefore, this study focuses on evaluating four distinct variations of KNN algorithms. The study aims to assess IPS performance metrics, including accuracy, precision, and computational cost, using four different KNN algorithms in a campus environment. The primary research questions are:

• How do different KNN algorithm variations perform on campus?
• Which Wi-Fi fingerprinting algorithm shows optimal performance in this setting?

Note: Please read full manuscript in the files.