Traditional K-means Clustering Research Articles

CAVE: Cerebral artery–vein segmentation in digital subtraction angiography

Cerebral X-ray digital subtraction angiography (DSA) is a widely used imaging technique in patients with neurovascular disease, allowing for vessel and flow visualization with high spatio-temporal resolution. Automatic artery–vein segmentation in DSA plays a fundamental role in vascular analysis with quantitative biomarker extraction, facilitating a wide range of clinical applications. The widely adopted U-Net applied on static DSA frames often struggles with disentangling vessels from subtraction artifacts. Further, it falls short in effectively separating arteries and veins as it disregards the temporal perspectives inherent in DSA. To address these limitations, we propose to simultaneously leverage spatial vasculature and temporal cerebral flow characteristics to segment arteries and veins in DSA. The proposed network, coined CAVE, encodes a 2D+time DSA series using spatial modules, aggregates all the features using temporal modules, and decodes it into 2D segmentation maps. On a large multi-center clinical dataset, CAVE achieves a vessel segmentation Dice of 0.84 (±0.04) and an artery–vein segmentation Dice of 0.79 (±0.06). CAVE surpasses traditional Frangi-based k-means clustering (P < 0.001) and U-Net (P < 0.001) by a significant margin, demonstrating the advantages of harvesting spatio-temporal features. This study represents the first investigation into automatic artery–vein segmentation in DSA using deep learning. The code is publicly available at https://github.com/RuishengSu/CAVE_DSA.

Fuzzy C-Means clustering for physical model calibration and 7-day, 10-year low flow estimation in ungaged basins: comparisons to traditional, statistical estimates

In the northeast U.S., resource managers commonly apply 7-day, 10-year (7Q10) low flow estimates for protecting aquatic species in streams. In this paper, the efficacy of process-based hydrologic models is evaluated for estimating 7Q10s compared to the United States Geological Survey's (USGS) widely applied web-application StreamStats, which uses traditional statistical regression equations for estimating extreme flows. To generate the process-based estimates, the USGS's National Hydrologic Modeling (NHM-PRMS) framework (which relies on traditional rainfall-runoff modeling) is applied with 36 years of forcings from the Daymet climate dataset to a representative sample of ninety-four unimpaired gages in the Northeast and Mid-Atlantic U.S. The rainfall-runoff models are calibrated to the measured streamflow at each gage using the recommended NHM-PRMS calibration procedure and evaluated using Kling-Gupta Efficiency (KGE) for daily streamflow estimation. To evaluate the 7Q10 estimates made by the rainfall-runoff models compared to StreamStats, a multitude of error metrics are applied, including median relative bias (cfs/cfs), Root Mean Square Error (RMSE) (cfs), Relative RMSE (RRMSE) (cfs/cfs), and Unit-Area RMSE (UA-RMSE) (cfs/mi2). The calibrated rainfall-runoff models display both improved daily streamflow estimation (median KGE improving from 0.30 to 0.52) and 7Q10 estimation (smaller median relative bias, RMSE, RRMSE, and UA-RMSE, especially for basins larger than 100 mi2). The success of calibration is extended to ungaged locations using the machine learning algorithm Fuzzy C-Means (FCM) clustering, finding that traditional K-Means clustering (FCM clustering with no fuzzification factor) is the preferred method for model regionalization based on (1) Silhouette Analysis, (2) daily streamflow KGE, and (3) 7Q10 error metrics. The optimal rainfall-runoff models created with clustering show improvement for daily streamflow estimation (a median KGE of 0.48, only slightly below that of the calibrated models at 0.52); however, these models display similar error metrics for 7Q10 estimation compared to the uncalibrated models, neither of which provide improved error compared to the statistical estimates. Results suggest that the rainfall-runoff models calibrated to measured streamflow data provide the best 7Q10 estimation in terms of all error metrics except median relative bias, but for all models applicable to ungaged locations, the statistical estimates from StreamStats display the lowest error metrics in every category.

Research on Railway Obstacle Detection Method Based on Developed Euclidean Clustering

To prevent the problem of safety accidents caused by the intrusion of obstacles into railway clearance, this paper proposes an obstacle detection method based on Light Detection and Ranging (LiDAR) to obtain and process rich three-dimensional (3D) information and depth information of the railway scene. The method first preprocesses the point cloud of the railway scenario collected by LiDAR to divide a basic area containing the rails. Then, the method divides the roadbed plane and fits the rails with the random sample consensus (RANSAC) algorithm, dividing the detection area according to the position of the rails. To address the issue of over or under-segmentation in the traditional Euclidean clustering method, which is due to sparser point clouds the farther the object is from the LiDAR, this paper improves the conventional Euclidean clustering. It introduces an adaptive distance threshold to categorize obstacles. Finally, compared with traditional Euclidean clustering, K-means clustering, and density-based spatial clustering of applications with noise (DBSCAN) clustering, the improved Euclidean cluster has achieved better results in terms of computing time and segmentation accuracy. Experimental results show the ability of the method to detect railway obstacles successfully.

Interpretable deep learning translation of GWAS and multi-omics findings to identify pathobiology and drug repurposing in Alzheimer's disease.

Translating human genetic findings (genome-wide association studies [GWAS]) to pathobiology and therapeutic discovery remains a major challenge for Alzheimer's disease (AD). We present a network topology-based deep learning framework to identify disease-associated genes (NETTAG). We leverage non-coding GWAS loci effects on quantitative trait loci, enhancers and CpG islands, promoter regions, open chromatin, and promoter flanking regions under the protein-protein interactome. Via NETTAG, we identified 156 AD-risk genes enriched in druggable targets. Combining network-based prediction and retrospective case-control observations with 10 million individuals, we identified that usage of four drugs (ibuprofen, gemfibrozil, cholecalciferol, and ceftriaxone) is associated with reduced likelihood of AD incidence. Gemfibrozil (an approved lipid regulator) is significantly associated with 43% reduced risk of AD compared with simvastatin using an active-comparator design (95% confidence interval 0.51-0.63, p<0.0001). In summary, NETTAG offers a deep learning methodology that utilizes GWAS and multi-genomic findings to identify pathobiology and drug repurposing in AD.

Analysis of the Intelligent Tourism Route Planning Scheme Based on the Cluster Analysis Algorithm.

In view of the problems of the traditional cluster analysis algorithm such as strong dependence on the initial cluster center, the traditional k-means cluster analysis algorithm is improved and the experiment proves that the improved algorithm has a better clustering effect; in view of the problems of the traditional tourism route planning, the improved k-means cluster analysis algorithm is applied to the intelligent tourism route planning scheme design and an intelligent tourism planning scheme based on the cluster analysis algorithm is proposed; the tourists' preference metric is fully considered, and the experimental results show that the scheme has certain reasonableness and reference value.

Research on Driving Conditions and Fuel Consumption of Improved K-means Clustering Algorithm

Abstract In order to solve the problem that the initial center of traditional clustering algorithm is easy to fall into local optimum and time-consuming. An improved combination optimization algorithm of principal component analysis and weighted K-means clustering is proposed. The algorithm introduces the maximum and minimum distance, weighted Euclidean distance, starting from the mean sum of the distances of the remaining clustering points, avoiding the influence of outliers and edge data. The proportion method is used to improve the principal component, and the characteristic influence factor obtained is used as the initial characteristic weight to construct a weighted Euclidean distance metric. According to the influence factors of feature contribution rate on clustering, a clustering method of feature weight influence factors is proposed. The representative feature factors are selected to highlight the clustering effect. Finally, the driving cycle of automobile is synthesized and the instantaneous fuel consumption is analyzed. The results show that: the difference value of speed acceleration joint distribution of the proposed method is only 1.05%, which saves 44.2% of the time compared with the traditional K-means clustering, and the driving cycle fitting degree is high, which can reflect the actual vehicle operation characteristics and fuel consumption.

Target Detection of Forward-Looking Sonar Image Based on Improved YOLOv5

Forward-looking sonar is a commonly used underwater detection device at present, but the detection accuracy is poor due to the complex underwater environment, small target highlight area and fuzzy feature details. Therefore, this paper proposes a forward sonar image target detection model based on You Only Look Once Version 5 (YOLOv5) network using transfer learning method. First, the YOLOv5 network is pretrained with COCO data set. Then the pre-training model is fine-tuned according to the training set of forward-looking sonar images. Before fine-tuning, the traditional k-means clustering is improved. The intersection over union ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$IoU$ </tex-math></inline-formula> ) value is used as the distance function to cluster the labeling information of the training set of the forward-looking sonar image. The results of clustering serve as the initial anchor frame of the training network. This operation greatly improves the detection speed. Second, due to the characteristics of weak echo intensity and small target area of forward-looking sonar image, an improved feature extraction method of CoordConv was proposed to give corresponding coordinate information to high-level features which improves the accuracy of network detection regression. Finally, the fine-tuned network is used to detect the target in the forward-looking sonar image. The experimental results show that the improved model based on YOLOv5 network is superior to the original YOLOv5 network and other popular deep neural networks for target detection in the forward-looking sonar image, which has a reference significance for underwater target detection. The CoordConv-YOLOv5 network based on transfer learning proposed in this paper shows the best performance in both detection accuracy and detection speed. Detection accuracy mAP@0.5:0.95 can reach 56.95%, and detection speed can reach 9ms.

On K-Means Clustering with IVIF Datasets for Post-COVID-19 Recovery Efforts

The recovery efforts of the tourism and hospitality sector are compromised by the emergence of COVID-19 variants that can escape vaccines. Thus, maintaining non-pharmaceutical measures amidst massive vaccine rollouts is still relevant. The previous works which categorize tourist sites and restaurants according to the perceived degree of tourists’ and customers’ exposure to COVID-19 are deemed relevant for sectoral recovery. Due to the subjectivity of predetermining categories, along with the failure of capturing vagueness and uncertainty in the evaluation process, this work explores the use k-means clustering with dataset values expressed as interval-valued intuitionistic fuzzy sets. In addition, the proposed method allows for the incorporation of criteria (or attribute) weights into the dataset, often not considered in traditional k-means clustering but relevant in clustering problems with attributes having varying priorities. Two previously reported case studies were analyzed to demonstrate the proposed approach, and comparative and sensitivity analyses were performed. Results show that the priorities of the criteria in evaluating tourist sites remain the same. However, in evaluating restaurants, customers put emphasis on the physical characteristics of the restaurants. The proposed approach assigns 12, 15, and eight sites to the “low exposure”, “moderate exposure”, and “high exposure” cluster, respectively, each with distinct characteristics. On the other hand, 16 restaurants are assigned “low exposure”, 16 to “moderate exposure”, and eight to “high exposure” clusters, also with distinct characteristics. The characteristics described in the clusters offer meaningful insights for sectoral recovery efforts. Findings also show that the proposed approach is robust to small parameter changes. Although idiosyncrasies exist in the results of both case studies, considering the characteristics of the resulting clusters, tourists or customers could evaluate any tourist site or restaurant according to their perceived exposure to COVID-19.

Price transmission in European fish markets

ABSTRACT We investigate price transmission in European fisheries markets. To start, we identify clusters of fish species both cross-regionally and within countries. A major issue in the clustering exercise is missing data due to reporting issues and seasonality in landed fish catches. To handle this we implement k-POD clustering which unlike traditional k-means clustering is able to account for missing data in clustering. Next, we move on to our primary goal of investigating price transmission through modelling price volatility spillovers. A missing value VAR framework is applied to identify directional spillovers among fish prices within clusters. We show that the directional volatility spillover effect is more prominent within each country for different species clusters than that for species cross-regionally. This suggests price transmission within localized markets for fish, rather than cross-regional markets for individual species. Although there are some neighbour country effects and some species that appear to be priced cross-regionally. Our study is the first to explore these multiple fish pricing dynamics, particularly taking account of the important issue of missing data which is a notable feature of fish pricing data.

Feature Selection and Mislabeled Waveform Correction for Water–Land Discrimination Using Airborne Infrared Laser

The discrimination of water–land waveforms is a critical step in the processing of airborne topobathy LiDAR data. Waveform features, such as the amplitudes of the infrared (IR) laser waveforms of airborne LiDAR, have been used in identifying water–land interfaces in coastal waters through waveform clustering. However, water–land discrimination using other IR waveform features, such as full width at half maximum, area, width, and combinations of different features, has not been evaluated and compared with other methods. Furthermore, false alarms often occur when water–land discrimination in coastal areas is conducted using IR laser waveforms because of environmental factors. This study provides an optimal feature for water–land discrimination using an IR laser by comparing the performance of different waveform features and proposes a dual-clustering method integrating K-means and density-based spatial clustering applications with noise algorithms to improve the accuracy of water–land discrimination through the clustering of waveform features and positions of IR laser spot centers. The proposed method is used for practical measurement with Optech Coastal Zone Mapping and Imaging LiDAR. Results show that waveform amplitude is the optimal feature for water–land discrimination using IR laser waveforms among the researched features. The proposed dual-clustering method can correct mislabeled water or land waveforms and reduce the number of mislabeled waveforms by 48% with respect to the number obtained through traditional K-means clustering. Water–land discrimination using IR waveform amplitude and the proposed dual-clustering method can reach an overall accuracy of 99.730%. The amplitudes of IR laser waveform and the proposed dual-clustering method are recommended for water–land discrimination in coastal and inland waters because of the high accuracy, resolution, and automation of the methods.

An improved K-means clustering algorithm for global earthquake catalogs and earthquake magnitude prediction

The occurrences of earthquakes have no any explicit warning, and earthquake magnitude prediction is still extremely challenging now. Therefore, this study proposes a seismic prediction model based on clustering of global earthquake data. First, an improved K-means clustering algorithm for global earthquake catalogs is proposed. Traditional K-means clustering has several limitations, i.e., the number of clusters needs to be initialized, the initial cluster centers are arbitrarily selected, and there is currently no magnitude parameter in the K-means clustering algorithm. To improve the algorithm, this study employs the space–time–magnitude (STM) distance and then proposes a maximum–minimum STM distance for the selection of the initial cluster centers. Additionally, the sum of squares error, Davies–Bouldin index, Calinski–Harabasz index, and silhouette coefficient are applied to determine the number of clusters. Subsequently, a seismic prediction model based on the clustering result combined with an artificial neural network is presented. Application of the improved clustering algorithm to the global seismic catalog from 1900–2019 obtained from the United States Geological Survey reveals better clustering accuracy than the traditional K-means algorithm and is also effective for seismic risk analysis in the local region. Furthermore, the seismic prediction model based on the clustering result also exhibits good performance, which has practical significance and reference value for future predictions of earthquake magnitude.

MapReduce distributed parallel computing framework for diagnosis and treatment of knee joint Kashin-Beck disease

To improve the accuracy and computational efficiency of the MapReduce distributed parallel computing framework, thereby mining the diagnosis and treatment data of Kashin-Beck Disease (KBD) of the knee joint. Based on the shortcomings of the traditional K-means Clustering Algorithm (KCA), a simplified method for distance calculation was proposed. The Manhattan distance was used instead of Euclidean distance. Further improvement strategies were proposed to implement and compare KCA of MapReduce (MR-KCA) and Improved MR-KCA (IMR-KCA). With the same data, the sum of squared errors of MR-KCA and IMR-KCA decreased with the increase in the number of center points. Compared with MR-KCA, the quality of IMR-KCA was higher, and their difference was especially evident at 8 GB data capacity. The total execution time of both MR-KCA and IMR-KCA increased with the increase in the number of center points. Compared to MR-KCA, the total execution time of IMR-KCA was significantly reduced, especially when the data capacity was 8 GB. When the number of center points was 5000, IMR-KCA could reduce the total execution time by 50%. Through experiments, IMR-KCA was proved to better present the diagnosis and treatment data of patients with knee joint KBD. The scalability rates of MR-KCA and IMR-KCA decreased as the number of nodes increased, but the scalability rates of both algorithms could be maintained above 0.80, which had better scalability. Compared with MR-KCA, IMR-KCA had significantly higher scalability. The IMR-KCA proposed in this study had high accuracy and computing efficiency, which could be used in the visualization of KBD diagnosis and treatment.

Slope-Based Shape Cluster Method for Smart Metering Load Profiles

Cluster analysis is used to study the group of load profiles from smart meters to improve the operability in distribution network. The traditional K-means clustering analysis method employs Euclidean distance as similarity measurement, which is insufficient in reflecting the shape similarities of load profiles. In this letter, we propose a novel shape cluster method based on the segmented slope of load profiles. Compared with traditional K-means and two improved algorithms, the proposed method can improve the clustering accuracy and efficiency by capturing the shape features of smart metering load profiles.

Optimization of geochemical anomaly detection using a novel genetic K-means clustering (GKMC) algorithm

Due to the complicated characteristics of regional geochemical data from stream sediments as a result of the complexity of geological features, detection of multi-elemental geochemical footprints of mineral deposits of interest is a challenging task. As a way to address this, a hybrid genetic algorithm-based technique, namely genetic K-means clustering (GKMC) algorithm, is proposed here for optimum delineation of multi-elemental patterns (both anomaly and background) in stream sediment geochemical data. To do so, factor analysis and sample catchment basin modeling were coupled with GKMC and traditional K-means clustering (TKMC) methods for identification of anomalous multi-elemental geochemical footprints of deposits of porphyry copper and skarn copper in the 1:100,000 scale Varzaghan map sheet, northwest Iran. Based on higher prediction rate, it can be inferred that the model derived by GKMC is superior to the one derived by TKMC. In addition, the strong anomaly classes of the GKMC and TKMC models predict, respectively, ~83% and ~66% of the porphyry/skarn Cu deposits in ~22% and ~36% of the study district. Thus, the geochemical anomaly targets derived by the GKMC method are more reliable than those generated by the TKMC method. This revealed that the GKMC algorithm is an efficient and robust tool for recognizing multi-element geochemical anomalies for mineral exploration.

Three-way clustering method for incomplete information system based on set-pair analysis

Traditional clustering algorithms clearly assign uncertain information into a single cluster, which does not fully indicate that a cluster may not have a clear boundary. For a large number of missing data, the traditional clustering method cannot achieve a good clustering effect on these datasets. Therefore, the idea of three-way decision is introduced into the traditional k-means clustering, as a result, the knowledge of set-pair information granule be combined. This paper presents a three-way clustering method which can process missing values effectively. First, for missing values, the granularity corresponding to missing values are recorded as the degree of difference. Next, the algorithm is going to establish the distance between the samples and the clustering centers according to the set-pair theory. All samples are assigned into clusters according to the size of the distance, and the clustering results with three-way are formed, which are positive region, boundary region and negative region, which improves the structure of clustering results. The samples of positive region certainly belong to this cluster; the samples of boundary region may belong to this cluster; the samples of negative region don’t belong to this cluster; and the clustering results are represented by the three regions together. Finally, the validity of the algorithm is verified by UCI dataset great work.

Approach to image segmentation based on interval neutrosophic set

As a generalization of the fuzzy set and intuitionistic fuzzy set, the neutrosophic set (NS) have been developed to represent uncertain, imprecise, incomplete and inconsistent information existing in the real world. Now the interval neutrosophic set (INS) which is an expansion of the neutrosophic set have been proposed exactly to address issues with a set of numbers in the real unit interval, not just one specific number. After definition of concepts and operations, INS is applied to image segmentation. Images are converted to the INS domain, which is described using three membership interval sets: T, I and F. Then, in order to increase the contrast between membership and evaluate the indeterminacy, a fuzzy intensification for each element in the interval set is made and a score function in the INS is defined. Finally, the proposed method is employed to perform image segmentation using the traditional k-means clustering. The experimental results on a variety of images demonstrate that the proposed approach can segment different sorts of images. Especially, it can segment 'clean' images and images with various levels of noise.

Robust Dimension Reduction for Clustering With Local Adaptive Learning.

In pattern recognition and data mining, clustering is a classical technique to group matters of interest and has been widely employed to numerous applications. Among various clustering algorithms, K-means (KM) clustering is most popular for its simplicity and efficiency. However, with the rapid development of the social network, high-dimensional data are frequently generated, which poses a considerable challenge to the traditional KM clustering as the curse of dimensionality. In such scenarios, it is difficult to directly cluster such high-dimensional data that always contain redundant features and noises. Although the existing approaches try to solve this problem using joint subspace learning and KM clustering, there are still the following limitations: 1) the discriminative information in low-dimensional subspace is not well captured; 2) the intrinsic geometric information is seldom considered; and 3) the optimizing procedure of a discrete cluster indicator matrix is vulnerable to noises. In this paper, we propose a novel clustering model to cope with the above-mentioned challenges. Within the proposed model, discriminative information is adaptively explored by unifying local adaptive subspace learning and KM clustering. We extend the proposed model using a robust l2,1 -norm loss function, where the robust cluster centroid can be calculated in a weighted iterative procedure. We also explore and discuss the relationships between the proposed algorithm and several related studies. Extensive experiments on kinds of benchmark data sets demonstrate the advantage of the proposed model compared with the state-of-the-art clustering approaches.

Multidimensional Kernel Principal Component Analysis of False Alarms of Rapidly Intensifying Atlantic Tropical Cyclones

Previous work investigating rapid intensification (RI) predictability within Atlantic tropical cyclones (TCs) has revealed numerous challenges in identifying parameters useful in predicting RI. In particular, false alarm RI forecasts are notably high, driving down forecast skill despite higher RI detection rates. RI forecast improvement will be achieved by identifying spatial patterns distinct to RI false alarm cases that help distinguish them from correctly forecast RI events. A previously developed machine learning ensemble was used to hindcast on a database of Atlantic TCs spanning 1999 – 2016 to identify false alarm RI forecasts, as well as correct RI forecasts. Each set of cases (correct and false alarm) was used to develop kernel principal component (KPC) based composites of important meteorological features within RI TCs to facilitate meteorological comparisons against correctly forecast events. Optimality was determined using a simple metric based on a cluster analysis silhouette and compared against a baseline of traditional k-means cluster analysis, hierarchical analysis, and rotated PC analysis. Once optimal cluster configurations were identified for each meteorological feature, events within each cluster were averaged to yield composites, which in turn were used to facilitate meteorological comparisons between false alarm and correctly forecast RI events.

Domain-specific market segmentation using a latent class mixture modelling approach and wine-related lifestyle (WRL) algorithm

PurposeSince the publication of Van Raaij and Verhallen’s seminal work in European Journal of Marketing in 1994, identifying the domain-specific market segmentation approach as one of the most feasible for segmenting markets, there has been surprisingly limited development in this field, with the food domain as the only exception. This study aims to develop a methodological approach using latent class mixture modelling as contribution in the domain-specific market segmentation field.Design/methodology/approachThis study captures the AIO lifestyle perspective using a domain-specific 80-item algorithm which has the wine (product) domain as its focus. A sample size of 811 consumers is used from data collected by means of the CATI approach.FindingsThe authors use four criteria for model selection: comparison of the Bayesian information criterion (BIC) statistic, comparison of classification error, verification of the interpretation of the derived segments and, finally, use of the conditional bootstrap procedure to test whether the selected model provides a significant improvement over the previous model. The five-segment model option yields a minimum BIC, the classification error measure is minimal and is easier to interpret than the other models. Segment descriptions for the five identified lifestyle-based segments are developed.Research limitations/implicationsSegmentation by traditional k-means clustering has proven to be less useful than the more innovative alternative of mixture regression modelling; therefore, the authors identify segments in the market on the basis of individuals’ domain-specific lifestyle characteristics using a latent class mixture modelling approach.Practical implicationsFollowing the attainment of a clear and robust market segmentation structure, the simultaneous analysis of the lifestyles, demographics and behaviours of consumers as nexus of the domain-specific segmentation approach, provides rich and valid information accurately informing the market segment descriptions.Originality/valueThe authors make a substantive contribution by developing a methodological approach using latent class mixture modelling; the first of its kind in the area of domain-specific segmentation. Next, they use the discriminant and/or predictive validity of the 80-scale items to predict cluster membership using the WRL algorithm. Finally, the authors describe the identified market segments in detail and outline the practical implications.

Business Process Modeling Abstraction Based on Semi-Supervised Clustering Analysis

The most prominent Business Process Model Abstraction (BPMA) use case is the construction of the process “quick view” for rapidly comprehending a complex process. Some researchers propose process abstraction methods to aggregate the activities on the basis of their semantic similarity. One important clustering technique used in these methods is traditional k-means cluster analysis which so far is an unsupervised process without any priori information, and most of the techniques aggregate the activities only according to business semantics without considering the requirement of an order-preserving model transformation. The paper proposes a BPMA method based on semi-supervised clustering which chooses the initial clusters based on the refined process structure tree and designs constraints by combining the control flow consistency of the process and the semantic similarity of the activities to guide the clustering process. To be more precise, the constraint function is discovered by mining from a process model collection enriched with subprocess relations. The proposed method is validated by applying it to a process model repository in use. In an experimental validation, the proposed method is compared to the traditional k-means clustering (parameterized with randomly chosen initial clusters and an only semantics-based distance measure), showing that the approach closely approximates the decisions of the involved modelers to cluster activities. As such, the paper contributes to the development of modeling support for effective process model abstraction, facilitating the use of business process models in practice.