Abdulmalik Shehu Yaro

Traditional clustering algorithms in fingerprint-based localization often struggle with outliers, overlapping clusters, and irregular RSS variations in fingerprint databases, which reduces clustering accuracy. To address these issues, this study proposes a hierarchical hybrid approach, the closest access point–medoids (CAP-medoids) algorithm, which combines the closest access point (CAP) method with k-medoids clustering. The CAP algorithm generates initial clusters based on the strongest received signal strength (RSS) from nearby wireless access points (APs), while k-medoids refines clusters by selecting actual fingerprint vectors as cluster centers, improving robustness against noise and irregular RSS variations. The algorithm was evaluated on four publicly available fingerprint databases of varying size and density. Performance was assessed using Euclidean, Manhattan, and cosine similarity distances as similarity metrics, with silhouette scores and Davies Bouldin (DB) indices as clustering performance metrics. Results show that the CAP-medoids algorithm consistently produces more compact and well-separated clusters than standard k-medoids in small databases, with silhouette scores increasing up to 75% and DB indices decreasing up to 63%. For larger, high-density databases, performance declines, indicating sensitivity to database size. Comparisons with other hybrid algorithms, including CAP+k-means++ and k-density-based spatial clustering of applications with noise (k-DBSCAN) algorithms, confirm its overall robustness and adaptability.

In fingerprinting-based localization systems, collecting time-series received signal strength (RSS) observations and averaging them is a common method for dealing with RSS fluctuation during the creation of the fingerprint database. However, in practice, these observations contain outliers caused by multipath propagation, interference, and hardware instability, which degrade the effectiveness of the averaging process. The modified Z-score method is widely adopted for outlier detection due to its simplicity and moderate accuracy. However, its performance depends strongly on the choice of scale estimator used to quantify data dispersion. A systematic evaluation of robust scale estimators under varying RSS observation sizes, with emphasis on detection reliability and computational cost, remains limited. This study evaluates six robust scale estimators within the modified Z-score framework, namely median absolute deviation (MAD), Sn, Qn, interquartile range (IQR), trimmed standard deviation (TSD), and Winsorized standard deviation (WSD). The goal is to identify a scale estimator that ensures reliable outlier detection while maintaining low to moderate computational cost for practical deployment in fingerprinting systems. Results show that, in terms of outlier detection performance, the IQR and Sn estimators achieve the lowest false detection rates across different observation sizes. In terms of computational efficiency, the MAD estimator provides the lowest processing time based on empirical runtime measurements. When both detection performance and computational efficiency are considered, the IQR estimator offers the most practical solution for preprocessing RSS time-series observations in fingerprint database creation for fingerprinting-based systems.

Because of received signal strength (RSS) variability in indoor environments, multiple observations are often collected sequentially over time to reduce short‐term fluctuations. However, these time‐series observations may contain local outliers caused by measurement disturbances and environmental variability. The local outlier factor (LOF) algorithm can detect such outliers, but its performance depends on the distance metric, neighborhood size, and LOF threshold. Existing studies provide limited comparative evaluation of distance metrics for LOF‐based outlier detection in time‐series RSS data. This study addresses this gap by systematically evaluating four distance metrics—namely, absolute, squared, relative, and Canberra—within the LOF framework for time‐series RSS outlier detection. Two RSS time‐series datasets were evaluated using neighborhood sizes of k = 10, 20, and 30 and LOF thresholds of T = 1.0, 1.5, and 2.0. Performance was assessed using receiver operating characteristic area under the curve (ROC‐AUC), precision‐recall area under the curve (PR‐AUC), precision, recall, and F1 score, with physically defined RSS bounds used to establish reference outliers. Dataset #1 contained 14 reference outliers, whereas Dataset #2 contained 175. The results show that distance metric selection affects threshold‐based classification performance, although its effect varies across datasets and parameter settings. For Dataset #1, squared distance achieved perfect precision, recall, and F1 score at k = 10 and T = 1.5. For Dataset #2, squared distance achieved the highest F1 score of 0.2927 at k = 30 and T = 1.5, whereas Canberra distance achieved the highest ROC‐AUC of 0.5812. The findings demonstrate that distance metric selection is an important consideration when configuring LOF for RSS outlier detection.

The projected clustering (PROCLUS) algorithm is a subspace clustering algorithm based on the k-medoids clustering approach. It is designed to address the challenges of irrelevant received signal strength (RSS) measurements in fingerprint vectors by focusing on meaningful subsets of RSS measurements from wireless APs, known as subspaces. Despite its robustness, its performance heavily depends on the chosen similarity metric, with Euclidean and Manhattan distances being the most common. While many researchers focus on modifying the algorithm to enhance performance, the impact of similarity metrics on clustering performance is often overlooked, despite its critical role in determining accuracy. As such, this paper evaluates the clustering performance of the PROCLUS algorithm using five distance-based similarity metrics—Euclidean, Manhattan, Cosine similarity, Canberra, and Chebyshev—across six experimentally generated fingerprint databases. It aims to identify the best similarity metric for maximizing clustering performance for each of the six fingerprint databases, with silhouette scores used as the clustering performance metric. Simulation results show that Cosine similarity is the most effective metric with the PROCLUS algorithm. It consistently produces clusters with the highest silhouette scores, all well above the 0.25 threshold and were 24% to 136% higher than the scores achieved using the other distance-based metrics across the tested fingerprint databases. The Canberra distance performed variably, while Euclidean and Manhattan distances were less reliable. The Chebyshev distance consistently underperformed in all the databases considered. The findings in this paper highlight the importance of choosing the appropriate similarity metric to perform clustering operations with the PROCLUS algorithm.

The correlation-based context similarity coefficient (CSC) metric is a pattern-based fingerprint similarity metric that has gained interest in fingerprint database clustering operations. The performance of the traditional distance-based metric in fingerprint vector similarity determination has been known to be influenced by the size of the fingerprint vector. However, as for the correlation-based CSC metric, there is no comprehensive research on how fingerprint vector size affects its performance in similarity determination and subsequently clustering performance. As such, this paper examines the impact of fingerprint vector size on the similarity determination performance of the correlation-based CSC metric with the k-medoids algorithm employed for clustering. The analysis is performed across four synthetic and two experimentally generated fingerprint databases with varying fingerprint vector sizes. The impact analysis is carried out with the clustering algorithm set to generate K ∈ [3, 5] clusters. Additionally, the results are compared against three distance-based metrics: squared Euclidean, Manhattan, and cosine. With silhouette score as the clustering performance metric, the simulation result shows that the size of the fingerprint vector influences the similarity determination performance of the correlation-based CSC metric. Additionally, the number of clusters in which the clustering algorithm is set to generate also contributes to how the correlation-based CSC metric performs in similarity determination. The similarity determination time complexity of the correlation-based CSC metric increases with fingerprint vector size, making efficient clustering more challenging as fingerprint vector sizes increase. For optimal performance, the correlation-based CSC metric is recommended for us as a similarity metric on a database with a fingerprint vector size of N ≤ 4 and a clustering algorithm configured to generate no more than 3 clusters.

Doi: 10.28991/HIJ-2025-06-01-022 Full Text: PDF

Fingerprint database clustering and localization using k‐medoids and k‐nearest neighbour (k‐NN) algorithms respectively typically use distance‐based fingerprint similarity metrics, with their performances dependent on the type of distance metric used. This paper proposes employing a pattern‐based metric, the context similarity coefficient (CSC), for both algorithms instead of traditional distance‐based metrics. The CSC accounts for fingerprint behaviour and the non‐linear relationships among fingerprints during the similarity measurement. The performance of both algorithms with the CSC as the similarity metric is evaluated on four publicly available fingerprint databases, using position root mean square error (RMSE) and silhouette score as performance metrics. These results are compared to those of the same algorithms using five distance‐based metrics: Euclidean, square Euclidean, Manhattan, cosine, and Chebyshev distances. The k‐medoids algorithm with CSC shows moderate clustering performance compared to the five distance‐based metrics considered. However, when combined with the k‐NN algorithm also using CSC, it achieves the highest localization accuracy, with at least a 29% improvement in position RMSE across all four databases. The results indicate that while k‐medoids with CSC may not create well‐separated clusters, combining it with the k‐NN algorithm with CSC as its similarity metric significantly enhances localization accuracy compared to distance‐based metrics.

The affinity propagation clustering (APC) algorithm is popular for fingerprint database clustering because it can cluster without pre-defining the number of clusters. However, the clustering performance of the APC algorithm heavily depends on the chosen fingerprint similarity metric, with distance-based metrics being the most commonly used. Despite its popularity, the APC algorithm lacks comprehensive research on how distance-based metrics affect clustering performance. This emphasizes the need for a better understanding of how these metrics influence its clustering performance, particularly in fingerprint databases. This paper investigates the impact of various distance-based fingerprint similarity metrics on the clustering performance of the APC algorithm. It identifies the best fingerprint similarity metric for optimal clustering performance for a given fingerprint database. The analysis is conducted across five experimentally generated online fingerprint databases, utilizing seven distance-based metrics: Euclidean, squared Euclidean, Manhattan, Spearman, cosine, Canberra, and Chebyshev distances. Using the silhouette score as the performance metric, the simulation results indicate that structural characteristics of the fingerprint database, such as the distribution of fingerprint vectors, play a key role in selecting the best fingerprint similarity metric. However, Euclidean and Manhattan distances are generally the preferable choices for use as fingerprint similarity metrics for the APC algorithm across most fingerprint databases, regardless of their structural characteristics. It is recommended that other factors, such as computational intensity and the presence or absence of outliers, be considered alongside the structural characteristics of the fingerprint database when choosing the appropriate fingerprint similarity metric for maximum clustering performance.