Enhancing Indoor Localization Estimation Using RSS Similarity-Based k-Nearest Neighbors

The rapid growth of indoor positioning is revolutionizing our understanding of entity locations within indoor spaces. The fingerprintbased indoor localization method using Wi-Fi access points (APs) stands out for its minimal hardware requirements, making it one of the promising techniques in this domain. The k-nearest neighbors (k-NN) algorithm, a common machine learning (ML) approach, provides location estimations by pinpointing the k neighbors with the most similar representation values. However, conventional distance functions utilized in k-NN, including Euclidean distance and cosine similarity, prove insufficient in accurately identifying nearest neighbors based on the meaningful interpretation of received signals from APs. Thus, in this research, we propose a new distance function based on received signal strength (RSS) similarity that can be employed in tandem with k-NN to find the optimal nearest neighbors for real-time localization on a more consistent basis when compared to other distance functions. The experimental results of the collected dataset demonstrated a 1 to 3% improvement in the coefficient of determination (R²) score and a reduction in distance error by 6.5 to 10 inches, as determined from the mean absolute error (MAE).

Recommended citation: Benyamain Yacoob, Daniel Marku, Mina Maleki, and Shadi Banitaan. 2025. "Enhancing Indoor Localization Estimation Using RSS Similarity-Based k-Nearest Neighbors." In Proceedings of the 2024 6^th Asia Conference on Machine Learning and Computing (ACMLC 24). Association for Computing Machinery, New York, NY, USA, 74–79. https://doi.org/10.1145/3690771.3690792
Download Paper | Download Slides