Affinity Propagation Research Articles

Short-term prediction of integrated energy load aggregation using a bi-directional simple recurrent unit network with feature-temporal attention mechanism ensemble learning model

In the realm of economic scheduling and optimal operation within integrated energy systems, integrated energy load aggregation plays a pivotal role. Accurate prediction of integrated energy load aggregation enables the refined management of energy resources, thereby enhancing energy utilization efficiency. This paper merges the advantages of clustering algorithms, prediction algorithms and ensemble learning to propose a short-term prediction approach for integrated energy load aggregation. Firstly, the intrinsic distribution of energy consumption data is analyzed through Affinity Propagation (AP) clustering with multi-dimensional similarity. This analysis yields the division of all samples into clusters based on data characteristics. Subsequently, the Transfer entropy is harnessed to conduct correlation analysis within each cluster. For constructing predictive models for each individual cluster, a prediction algorithm based on the Feature-Temporal Attention-enhanced Bi-directional Simple Recurrent Unit (FTA-Bi-SRU) is employed. The Dynamic Weight Average (DWA)-based multi-task loss equilibrium and Multi-Term Adam (MTAdam) loss function are combined to enhance the efficiency of the forecasting model. Finally, a Light Gradient Boosting Machine (LightGBM) model with Bayesian optimization is employed to integrate the predictive outcomes of each individual cluster to generate the overall predictive outcomes for integrated energy load aggregation. The proposed methodology is experimented on real-world energy consumption data to ascertain the efficacy of the prediction method. The experimental results substantiate the superiority of the suggested forecasting approach.

Operational optimization of copper flotation process based on the weighted Gaussian process regression and index-oriented adaptive differential evolution algorithm

Concentrate copper grade (CCG) is one of the important production indicators of copper flotation processes, and keeping the CCG at the set value is of great significance to the economic benefit of copper flotation industrial processes. This paper addresses the fluctuation problem of CCG through an operational optimization method. Firstly, a density-based affinity propagationalgorithm is proposed so that more ideal working condition categories can be obtained for the complex raw ore properties. Next, a Bayesian network (BN) is applied to explore the relationship between the operational variables and the CCG. Based on the analysis results of BN, a weighted Gaussian process regression model is constructed to predict the CCG that a higher prediction accuracy can be obtained. To ensure the predicted CCG is close to the set value with a smaller magnitude of the operation adjustments and a smaller uncertainty of the prediction results, an index-oriented adaptive differential evolution (IOADE) algorithm is proposed, and the convergence performance of IOADE is superior to the traditional differential evolution and adaptive differential evolution methods. Finally, the effectiveness and feasibility of the proposed methods are verified by the experiments on a copper flotation industrial process.

A Novel Hybrid Protocol and Code Related Information Reconciliation Scheme for Physical Layer Secret Key Generation

Wireless networks are vulnerable to various attacks due to their open nature, making them susceptible to eavesdropping and other security threats. Eavesdropping attack takes place at the physical layer. Traditional wireless network security relies on cryptographic techniques to secure data transmissions. However, these techniques may not be suitable for all scenarios, especially in resource-constrained environments such as wireless sensor networks and adhoc networks. In these networks having limited power resources, generating cryptographic keys between mobile entities can be challenging. Also, the cryptographic keys are computationally complex and require key management infrastructure. Physical Layer Key Generation (PLKG) is an emerging solution to address these challenges. It establishes secure communication between two users by taking advantage of the wireless channel's inherent features. PLKG process involves channel probing, quantization, information reconciliation (IR) and privacy amplification to generate symmetric secret key. The researchers have used various PLKG techniques to get the secret key, sTop of Form
 till they face problems in the IR scheme to obtain symmetric keys between the users who share the same channel for communication. Both the code based and protocol based methods proposed in the literature have advantages and limitations related to their performance parameters such as information leakage, interaction delay and computation complexity. This research work proposes a novel IR mechanism that combines the protocol and code-based error correction methods to obtain reduced Bit Mismatch Rate (BMR), reduced information leakage, reduced interaction delay, and reduced computational time to enhance physical layer secret key's quality. In the proposed research work, the channel samples are generated using the Received Signal Strength (RSS) and Channel Impulse Response (CIR) parameters. These samples are quantized using Vector Quantization with Affinity Propagation Clustering (VQAPC) method to generate the preliminary key. The samples collected by the two users who wish to communicate, (for example Alice and Bob) will be different due to noise in the channel and hardware limitations. Hence their preliminary keys will be different. Removing this discrepancy between Bob's and Alice's initial keys, using novel Hybrid Protocol and Code related Information Reconciliation (HPC-IR) scheme to generate error corrected key, is the most important contribution of this research work. This key is further encoded by the MD5 hash function to generate a final secret key for exchanging information between two users over the wireless channel. It is observed that the proposed HPC-IR scheme achieves BMR of 19.4%, information leakage is 0.002, interaction delay is 0.001 seconds and computation time is 0.02 seconds.

Geometric Affinity Propagation for Clustering With Network Knowledge

Clustering data into meaningful subsets is a major task in scientific data analysis. To date, various strategies ranging from model-based approaches to data-driven schemes, have been devised for efficient and accurate clustering. One important class of clustering methods that is of a particular interest is the class of exemplar-based approaches. This interest primarily stems from the amount of compressed information encoded in these exemplars that effectively reflect the major characteristics of the respective clusters. Affinity propagation (AP) has proven to be a powerful exemplar-based approach that refines the set of optimal exemplars by iterative pairwise message updates. However, a critical limitation is its inability to capitalize on known networked relations between data points often available for various scientific datasets. To mitigate this shortcoming, we propose geometric-AP, a novel clustering algorithm that effectively extends AP to take advantage of the network topology. Geometric-AP obeys network constraints and uses max-sum belief propagation to leverage the available network topology for generating smooth clusters over the network. Extensive performance assessment reveals a significant enhancement in the quality of the clustering results when compared to benchmark clustering schemes. Especially, we demonstrate that geometric-AP performs extremely well even in cases where the original AP fails drastically.

CAD-CDN: Coronary Artery Disease Prediction Using Convolutional Neural Network with Modified Densenet

Atherosclerosis is a synonym for coronary artery disease (CAD), a non-communicable cardiovascular disease. Coronary artery disease, cancer, and tumour illness pose significant human risks. Predicting coronary artery disease (CAD) is a difficult and time-consuming task in the medical field. Early prediction is a virtuoso skill in the medical area, particularly in the cardiovascular sector. Prior research on developing early prediction models provided a grasp of modern strategies for detecting variance in medical imaging. Cardiovascular disease prevention may be accomplished with a diet plan established by the concerned physician after early diagnosis. We proposed a CAD-CDN framework for coronary artery disease prediction using a Convolutional neural network (CNN) with modified densenet. The datasets are collected from the Kaggle repository, and the data normalization has been done with Affinity propagation with an adaptive damping factor (APADF). The best features are selected using ACO with SA as the Hybrid method. Finally, the classification was done with CNN with modified Densenet. The experimental result has been done with various existing algorithms and proposed one. And the results have shown performance indicators including accuracy, precision, sensitivity, specificity, and measure value.

Undersea Water Objects Detection and Classification Using Optimized Region-Based Convolutional Neural Network

Underwater autonomous operation is becoming increasingly crucial to avoid the hazardous in the environment of high-pressure deep-sea due to the significance of underwater investigation. The most crucial piece of technology for underwater-based task is intelligent computer vision. In an underwater environment, underwater vision requires good image quality, and illumination with better classification of sea objects. This work presented a novel technique of Optimized Region-based Convolutional Neural Network (ORCNN). For this work, the Gaussian filter is used to remove the noise and enhance the image quality during pre-processing and the Improved Affinity Propagation Clustering (IAPC) model segments the objects. After that, the Region-based Convolutional Neural Network (RCNN) model classifies various objects such as urchins, seagrass, fishes, and rocks in which the RCNN parameters are tuned via a light spectrum optimizer algorithm (LSOA). can also input the segmentation prediction images and the labeled images into the discriminant convolutional network and improve the segmentation accuracy of underwater images by further enhancing the essential characteristics of learning data through the confrontation training of generators and discriminators. The experimental results demonstrate that higher performance is provided by the newly developed ORCNN -Net predictive model when compared to other comparative algorithms while considering the negative and positive metrics.

Improving the Efficiency of Modern Warehouses Using Smart Battery Placement

In the ever-evolving landscape of warehousing, the integration of unmanned ground vehicles (UGVs) has profoundly revolutionized operational efficiency. Despite this advancement, a key determinant of UGV productivity remains its energy management and battery placement strategies. While many studies explored optimizing the pathways within warehouses and determining ideal power station locales, there remains a gap in addressing the dynamic needs of energy-efficient UGVs operating in tandem. The current literature largely focuses on static designs, often overlooking the challenges of multi-UGV scenarios. This paper introduces a novel algorithm based on affinity propagation (AP) for smart battery and charging station placement in modern warehouses. The idea of the proposed algorithm is to divide the initial area into multiple sub-areas based on their traffic, and then identify the optimal battery location within each sub-area. A salient feature of this algorithm is its adeptness at determining the most strategic battery station placements, emphasizing uninterrupted operations and minimized downtimes. Through extensive evaluations in a synthesized realistic setting, our results underscore the algorithm’s proficiency in devising enhanced solutions within feasible time constraints, paving the way for more energy-efficient and cohesive UGV-driven warehouse systems.

Clustering algorithm for experimental datasets using global sensitivity-based affinity propagation (GSAP)

To minimize the uncertainty of the parameters in combustion kinetics models, Bayesian methods are commonly used for uncertainty constraints based on experimental data. With the rapid and substantial growth of experimental data, using all the experimental data for optimization is not only redundant and time-consuming, but it could also lead to data consistency problems. In this work, the global sensitivity-based affinity propagation method (GSAP) is proposed to cluster experimental datasets and to select representative experimental conditions. Specifically, the global sensitivity coefficient is first obtained through an analysis to characterize the sources of uncertainty in the kinetic model under different experimental conditions. The similarity coefficient, which is defined based on the global sensitivity, measures the resemblance between two experimental conditions. By exchanging messages calculated from similarity, affinity propagation enables the experimental dataset to be automatically clustered into several classes without specifying the number of classes in advance. This method innovatively introduces the consideration of model and experimental uncertainty under different conditions to obtain better optimization results. The correctness and effectiveness of the method are validated through clustering and optimizing on a laminar flame speed dataset of common C0C4 fuels. The dataset consisting of 288 experimental conditions has been automatically clustered into 27 categories, and an exemplar of each category is given. These exemplary conditions reflect the dominant chemistry behind their cluster. At the same time, these conditions have larger model prediction uncertainty and smaller experimental uncertainty to provide better Bayesian constraints. The uncertainty of the model parameters after Bayesian optimization is effectively constrained. The average uncertainty of model predictions across the dataset is reduced from 30 % to 10 % using only 27 exemplar conditions for optimization. While selecting experimental data for model optimization, the clustering strategies provided by this method also, in turn, help understand its underlying chemical essence.

Improvement of motor imagery electroencephalogram decoding by iterative weighted Sparse-Group Lasso

Discriminative feature selection is vital for enhancing motor imagery decoding performance in electroencephalogram (EEG) signals. However, existing feature optimization methods have not sufficiently explored the intrinsic attribute distribution of features and their associations with the target class, which could result in spurious correlations between optimized features and class labels, yielding suboptimal performance. Therefore, this study proposed an iterative Weighted Sparse-Group Lasso (iWSGL) model for optimizing Common Spatial Pattern (CSP)-based high-dimensional features, thus further enhancing the decoding accuracy of motor imagery. Specifically, the affinity propagation (AP) clustering algorithm was utilized to adaptively partition the high-dimensional features into multiple groups based on the underlying relationships among them. To evaluate the significance of individual feature within each group and the overall significance of the groups themselves, a weight calculation method was proposed based on conditional entropy. With the weights and feature structural information, a weighted sparse regression model was devised within the iterative Sparse-Group Lasso (iSGL) framework to jointly optimize the CSP-based high-dimensional features. The performance of the proposed method was validated on three datasets using the support vector machine (SVM). The experimental results exhibited the exceptional superiority of the proposed method over the current CSP and its variants, demonstrating its remarkable performance. These findings imply that the proposed model can offer a novel optimization strategy for enhancing pattern recognition of brain intentions in Brain-Computer Interface (BCI) applications.

A spatio-functional logistics profile clustering analysis method for metropolitan areas

This paper proposes a framework to define a zoning procedure using clustering taking into account socio-economic, spatial and logistics intensity variables to support urban logistics planning and management decision making. The methodology is centered around comparing two dimension reduction algorithms (PCA and UMAP) and four clustering algorithms (k-means, affinity propagation, HDBSCAN, and SOM). This comparison is based on combinations of dimension reduction and clustering techniques, assessing the results for geographic coherence, patterns that are captured and the statistical validity of the clustering results. The variables used in the clustering are defined from socio-economic, geographic and demographic data issued from standard sources, and a logistics intensity estimation via freight trip generation (FTG models). Within its application to Lima, Peru, results show that the choice of the FTG model, the main logistics intensity variable, has a strong impact on the final composition of the logistics profile and also on ensuring a geographical sense of the clustering results. Finally, research, policy, and practical implications are discussed, as well as future research stemming from these results.

A progressive framework combining unsupervised and optimized supervised learning for debris flow susceptibility assessment

This research aims to improve the reliability of debris flow susceptibility (DFS) assessment, which is crucial for disaster prevention and mitigation in mountainous regions. A progressive framework was proposed and applied to Pinggu District, Beijing, China. First, 16 debris flow predisposing factors (DFPFs) were selected, and the slope structure and soil stability were incorporated to account for material sources. Watershed units instead of grid units were used to extract data. Then, the multi-collinearity among the factors was reduced by using variance inflation factors (VIF) and information gain (IG), and 13 DFPFs were retained. Three unsupervised learning algorithms (i.e., affinity propagation (AP), Gaussian mixture model (GMM) and self-organizing maps (SOM)) were used to optimize the sampling strategy of non-debris flow units. Subsequently, the occurrence probability of debris flows in each unit was predicted by using four supervised learning algorithms (i.e., logistic regression (LR), random forest (RF), adaptive boosting (ADAB) and extreme gradient boosting (XGB)). They were optimized by a state-of-the-art meta-modeling approach. Finally, the DFPF’s importance was ranked. The main contributions of our framework are establishing a high-quality data set and optimizing the prediction algorithms. The results show that the tree-based models perform well, and the boosting-based algorithms outperform the bagging-based algorithms. Supervised learning is more suitable for DFS assessment than unsupervised learning. Debris flows are most likely to occur on a dolomite consequent or diagonal slope with a relief amplitude above 540 m.

Double similarities weighted multi-instance learning kernel and its application

Multi-instance learning (MIL), as a special version of classification, focuses on labeled sets (bags) consisting of unlabeled instances and has drawn accumulative attention due to its significant importance in practical applications. However, most existing MIL methods just utilize partial information (bags or instances) of MIL data to construct the kernel function, resulting in deteriorated classification performance of MIL. In this paper, we propose a Double Similarities weighted Multi-Instance Learning (DSMIL) kernel framework, which utilizes the similarities of Bag-to-Bag (B2B) and Instance-to-Bag (I2B). In the proposed kernel framework, the similarities of B2B and I2B could be derived from the prototypes distance of inter-bag and similarity matrix of intra-bag, respectively, based on the affinity propagation (AP) clustering of the bag. Meanwhile, we give theoretical proof of the validity of the designed kernel function. Experimental results on benchmark and semi-synthetic datasets show that our proposed method obtains competitive classification performance and achieves robustness to parameters and noise.

B-125 Artificial Intelligence Application to Improve the Algorithm for Autoverification of Results of Biochemical Markers that Assess Thyroid Function

Abstract Background The results verification process is the most important control step of the post-analytical phase. It can be done in two ways, manually by a professional, or using autoverification systems. Appropriate results are verified and submitted to the Laboratory Information System (LIS). Corrective actions are initiated for discordant results prior to their submission to the LIS. Autoverification is a process where the laboratory tests results are published without any professional’s intervention. It is recommended that each laboratory creates the necessary cut-off values for autoverification, according to its population and its biochemical-medical criteria. In our laboratory, an autoverification algorithm is applied to evaluate thyroid function, which involves the measurement of 2 parameters: Thyrotropin (TSH) and Free Thyroxine (FT4). These parameters were chosen because they are the most used for characterizing thyroid function. The autoverification algorithm was defined for euthyroid, hypothyroid and hyperthyroid patients. Our professionals proposed a “safe range” for both parameters and we relate those parameters results for autoverification. We study the usefulness of artificial intelligence models by clustering non-supervised machine learning algorithms, to find behavioral patterns of not autoverified sample results, to support and even broaden the “safe ranges” established by our professionals. The goal was to improve the current autoverification algorithm by using artificial intelligence for thyroid function test results and to increase the autoverification percentage. Methods The autoverification process was carried out at TURNER Laboratories, Rosario, Argentina. The autoverification algorithm was designed according to the Laboratory and Clinical Standards Institute Guide AUTO10-A. The Atellica Data Manager (ADM) middleware was used to provide the connection between the Atellica Solution Immunoassay & Clinical Chemistry Analyzers and the LIS. ADM receives the results from the analyzers and processes them in consideration of the pre-analytical (age, sex, etc.), analytical (QC, flags, etc.) and post-analytical (Delta Check, etc.) conditions. It then applies the autoverification algorithm to send the results to the LIS. In the first semester of 2022, we used the autoverification algorithm with the existing “safe ranges” in 18 626 samples of outpatients between the ages of 16 and 60. On the other hand, we applied different Clustering algorithms to the samples that were not autoverified, to find behavior patterns. The ones we used are: KMeans, Fuzzy C-Means, DBSCAN, Agglomerative Clustering, Dendrogram and Affinity Propagation. We chose KMeans with k = 4. Our professionals thoroughly analyzed each of the clusters and defined “new safe ranges” for autoverification. In the second semester of 2022, 18 947 samples were analyzed and the autoverification algorithm was applied with the “safe ranges” and the “new safe ranges”. Results In the first semester of 2022, by applying the existing “safe ranges” we obtained an autoverification of 19.0%. We then simulated an autoverification with “safe ranges” and “new safe ranges” and obtained an autoverification of 71.5%. In the second semester, we applied autoverification with “safe ranges” and “new safe ranges” obtaining an autoverification of 72.7%. Conclusion The application of machine learning algorithms has significantly improved the level of autoverification of biochemical markers that evaluate thyroid function.

AP clustering algorithm for analysis of the impact of cold chain distribution center location on logistics costs

ABSTRACT The distribution center is a transit place for goods in the logistics network, used to achieve the distribution of goods. Compared with ordinary logistics, cold chain logistics has higher requirements for timeliness due to the low temperature or ultra-low temperature requirements of transport objects. Aiming at the problems of high cost and low efficiency of cold chain distribution center location, a new location model of cold chain distribution center is developed. The Affinity Propagation (AP) clustering algorithm is used to simplify location selection. And combine the binary semantics with entropy to reduce the subjectivity of the binary semantics in the process of experiment. The results show that the location selection using the research method is optimal, and the problem that multiple secondary distribution centers are the same retailer will not appear. The research method is more objective and scientific for the location of cold chain distribution centers.

Hierarchical Affinity Propagation for Reliable Integrated Multilayer V2X Networks

The integration of IEEE802.11p and 5G New Radio for the backbone transmission of V2X (vehicle-to-everything) technology has served as a novel type of intelligent transportation system with an efficient low-cost and high-coverage area. This integrated scheme implements a V2X end-to-end data delivery system in a multi-layer mechanism. However, the dynamic nature of vehicular networks leads to numerous network outages and frequent network renewal requests, which considerably diminishing the overall performance. Hence, efficient routing management is required to maximize reliable communication. Optimizing routing management is categorized as a non-linear problem with a very high computational load, making a practical solution very challenging. To cope with the complicated routing problem, we develop a low-complexity distributed algorithm based on a hierarchical affinity propagation framework. The proposed algorithm exchanges small-sized messages among local vehicles to determine data routing more efficiently. The distributive optimization of the proposed algorithm requires a small computational load for each node. The main advantage of the proposed algorithm is that each vehicle autonomously determines its best routing options without the help of any network coordinators. Extensive simulation results verify that the proposed algorithm outperforms conventional algorithms in terms of vehicle mobility, network scalability, the level of network overhead, and the algorithm convergence rate.

Population affinity propagation approach for points of dispensing location allocation

Population affinity propagation approach for points of dispensing location allocation

A cluster of blood-based protein biomarkers associated with decreased cerebral blood flow relates to future cardiovascular events in patients with cardiovascular disease

Biological processes underlying decreased cerebral blood flow (CBF) in patients with cardiovascular disease (CVD) are largely unknown. We hypothesized that identification of protein clusters associated with lower CBF in patients with CVD may explain underlying processes. In 428 participants (74% cardiovascular diseases; 26% reference participants) from the Heart-Brain Connection Study, we assessed the relationship between 92 plasma proteins from the Olink® cardiovascular III panel and normal-appearing grey matter CBF, using affinity propagation and hierarchical clustering algorithms, and generated a Biomarker Compound Score (BCS). The BCS was related to cardiovascular risk and observed cardiovascular events within 2-year follow-up using Spearman correlation and logistic regression. Thirteen proteins were associated with CBF (ρSpearman range: −0.10 to −0.19, pFDR-corrected <0.05), and formed one cluster. The cluster primarily reflected extracellular matrix organization processes. The BCS was higher in patients with CVD compared to reference participants (pFDR-corrected <0.05) and was associated with cardiovascular risk (ρSpearman 0.42, p < 0.001) and cardiovascular events (OR 2.05, p < 0.01). In conclusion, we identified a cluster of plasma proteins related to CBF, reflecting extracellular matrix organization processes, that is also related to future cardiovascular events in patients with CVD, representing potential targets to preserve CBF and mitigate cardiovascular risk in patients with CVD.

Clustering the Perceptions of Labelled Actors of the Vallée de la Gastronomie-France

Abstract Purpose Vallée de la Gastronomie-France (VG) is a new label-brand-network project applied by the territorial governments of three regions in France: Bourgogne-Franche-Comté, Auvergne-Rhône-Alpes, and Provence-Alpes-Côte-d’Azur. The purpose of this study was to investigate this project in terms of its business application and labelled actors’ perception. Design/methodology/approach The methodology was based on mixed methods. The qualitative methods included interviews, observations, and study of the job. At the same time, the databases from the Tourism Of-fices were structured according to data clustering methods such as Affinity propagation and DBSCAN. The results were obtained by using Affinity propagation. Findings The study reveals eight clusters of the VG’s labelled actors with a specific attitude toward aggregated attributes which have been selected to describe the project perception. After analysis from experts, they were classified as skeptical, undecided, or supportive toward the VG project. However, each of the clusters has its own distinctive perspective on collaboration, labelling, and the networking process. Practical implications Our findings allowed us to bring about a list of recommendations for the Tourism Offic-es’ managers, who are in charge of relationship making with the labelled actors, as well as for the regional struc-tures which are responsible for the VG’s further development. Social implications From the labelled actors’ point of view, the most valuable meaning was related to partici-pation in the project of the new brand destination, where wine and gastronomy activities are core competences. Originality/value Our paper contributes to the management of the VG by identifying its potential destination

Two-Phases Flying Process Based Cruise Time Minimization for Collecting Data in a UAV-Enabled Untraveled Wireless IoT Network

This work studies a practical scenario where the unmanned aerial vehicle (UAV) is dispatched from the data center to collect the data from the Internet of Things (IoT) devices and return to the data center. These IoT devices with different amounts of data and are sparsely located in an untraveled area. In this scenario, we consider minimizing the UAV's cruise time by jointly optimizing the clustering strategy of the IoT devices, the association strategy between the UAV and the IoT devices, the UAV's trajectory, and the bandwidth allocation strategy. Traditionally, the minimal cruise time is obtained by visiting all IoT devices one by one and this will lead to huge cost. To address this problem, affinity propagation (AP) algorithm is proposed to cluster all these IoT devices first of all. Then, to avoid visiting all devices one by one and realize parallel computation of data collection time, the entire UAV's flying process is divided into two phases, namely, the inter-cluster phase and the intra-cluster phase. For the inter-cluster phase, the minimal cruise time is formulated to minimize total distance among clusters using the traveling salesman problem with a neighborhood (TSPN)-based algorithm. Subsequently, for the intra-cluster phase, we formulate the collecting time minimization problem in each cluster as a mix-integer non-convex optimization problem. To address these issues, we design a two-loop algorithm involving the bisection search, the altering optimization (AO), and the successive convex approximation (SCA) methods. Simulation results validate the effectiveness of our proposed scheme.

A Clustering Technique for Ultrawideband Channel Using Modified Affinity Propagation

In this study, we present an ultra-wideband clustering algorithm using affinity propagation (AP). Unlike typical AP algorithms, which define similarity as a negative squared Euclidean distance between data points, we define it to reflect the amplitude discontinuity observed in Saleh-Valenzuela models. The proposed algorithm was tested on simulated channel impulse response and propagation data measured in an indoor environment, and reasonable clustering results were observed.