Sequential Clustering Algorithm Research Articles

Prediction of wind energy location by parallel programming using MPI-based KMEANS clustering algorithm

ABSTRACT Renewable energy resources, like the power of the wind, are the essential sources of energy in today’s world. To keep down the greenhouse energy degasification and stop global warming, it is very important to predict the exact location where maximum wind energy is generated. The total amount of electrical power produced by a turbine relies more on the speed of the wind, pressure created by the wind, and weather conditions. The proposed method predicts the maximum power generated in a particular location using the current conditions of the weather, pressure generated through wind, and its speed. The speed of the wind and pressure of the wind are clustered using the Message Passing Interface (MPI) based KMEANS clustering algorithm. The system minimizes the amount of time required for clustering and it is done by MPI. Clustering is formed with the help Euclidean distance of each point. The wind data is collected and formed into three clusters such as low, medium, and high based on the speed of the wind. The parameters for evaluation are considered as the speed of the wind, and the direction of the wind are determined. The results show that the speed of the wind varies up to 25 km/s with the power of 2000 watts. The proposed method compares the execution time of sequential and MPI-based KMEANS clustering for different numbers of clusters. The sequential KMEANS clustering algorithm takes 3% to 7% more time compared to MPI-based KMEANS clustering algorithm. For the maximum cluster size of 5, the MPI-based KMEANS clustering algorithm produces the result in 0.42 s.

Improved Fingerprint-Based Localization Based on Sequential Hybridization of Clustering Algorithms

The localization accuracy of a fingerprint-based localization system is dependent on several factors, one of which is the accuracy and efficiency at which the fingerprint database is clustered. Most highly efficient and accurate clustering algorithms have high time-dependent computational complexity (CC), which tends to limit their practical applicability. A technique that has yet to be explored is the sequential hybridization of multiple low-time CC clustering algorithms to produce a single moderate-time CC clustering algorithm with high localization accuracy. As a result, this paper proposes a clustering algorithm with a moderate time CC that is based on the sequential hybridization of the closest access point (CAP) and improved k-means clustering algorithms. The performance of the proposed sequential hybrid clustering algorithm is determined and compared to the modified affinity propagation clustering (m-APC), fuzzy c-mean (FCM), and 2-CAP algorithms presented in earlier research works using four experimentally generated and publicly available fingerprint databases. The performance metrics considered for the comparisons are the position root mean square error (RMSE) and clustering time based on big O notation. The simulation results show that the proposed sequential hybrid clustering algorithm has improved localization accuracy with position RMSEs of about 54%, 77%, and 52%, respectively, higher than those of the m-APC, FCM, and 2-CAP algorithms. In terms of clustering time, it is 99% and 79% faster than the m-APC and FCM algorithms, respectively, but 90% slower than the 2-CAP algorithm. The results have shown that it is possible to develop a clustering algorithm that has a moderate clustering time with very high localization accuracy through sequential hybridization of multiple clustering algorithms that have a low clustering time with poor localization accuracy. Doi: 10.28991/ESJ-2024-08-02-02 Full Text: PDF

A deep inverse convolutional neural network-based semantic classification method for land cover remote sensing images

The imbalance of land cover categories is a common problem. Some categories appear less frequently in the image, while others may occupy the vast majority of the proportion. This imbalance can lead the classifier to tend to predict categories with higher frequency of occurrence, while the recognition effect on minority categories is poor. In view of the difficulty of land cover remote sensing image multi-target semantic classification, a semantic classification method of land cover remote sensing image based on depth deconvolution neural network is proposed. In this method, the land cover remote sensing image semantic segmentation algorithm based on depth deconvolution neural network is used to segment the land cover remote sensing image with multi-target semantic segmentation; Four semantic features of color, texture, shape and size in land cover remote sensing image are extracted by using the semantic feature extraction method of remote sensing image based on improved sequential clustering algorithm; The classification and recognition method of remote sensing image semantic features based on random forest algorithm is adopted to classify and identify four semantic feature types of land cover remote sensing image, and realize the semantic classification of land cover remote sensing image. The experimental results show that after this method classifies the multi-target semantic types of land cover remote sensing images, the average values of Dice similarity coefficient and Hausdorff distance are 0.9877 and 0.9911 respectively, which can accurately classify the multi-target semantic types of land cover remote sensing images.

A study on the clusterability of latent representations in image pipelines.

Latent representations are a necessary component of cognitive artificial intelligence (AI) systems. Here, we investigate the performance of various sequential clustering algorithms on latent representations generated by autoencoder and convolutional neural network (CNN) models. We also introduce a new algorithm, called Collage, which brings views and concepts into sequential clustering to bridge the gap with cognitive AI. The algorithm is designed to reduce memory requirements, numbers of operations (which translate into hardware clock cycles) and thus improve energy, speed and area performance of an accelerator for running said algorithm. Results show that plain autoencoders produce latent representations which have large inter-cluster overlaps. CNNs are shown to solve this problem, however introduce their own problems in the context of generalized cognitive pipelines.

Nonintrusive Load Identification for Industrial Users Integrated with LSQR and Sequential Leader Clustering

Nonintrusive load identification for industrial users can accurately acquire the operation of each load. However, it is a major challenge in the demand-side response due to the hardship of collection data for modelling, and high precision measuring equipment is required. Aiming at this situation, a nonintrusive load identification method is proposed, combining the least square QR (LSQR) with the sequential leader clustering algorithm. Firstly, regarding accurate depiction of industrial loads, some appropriate load feature indices of steady-state and transient processes are extracted, respectively. For steady-state processes, the active power, the reactive power, and the root mean square (RMS) current value are selected as the feature indices. In the case of transient processes, ten feature indices of three stages are employed: before, during, and after transient events, consisting of the duration of transient events, the RMS current value before and after transient events, the average value of active power before and after transient events, the average value of reactive power before and after transient events, the maximum RMS current value of transient events, etc. On this base, the LSQR algorithm is proposed to decompose unknown composite power to access the operation of various loads at steady-state. The sequential leader clustering algorithm is propounded to classify transient events of typical industrial loads and further identify which kind of loads had switched. Finally, to validate the effectiveness of the presented model, data of industrial loads from a concrete plant are collected, including blender, cement screw, sewage dump, and inclined belt conveyor, and simulation analysis is fulfilled. The results indicate that the model proposed can effectively achieve the nonintrusive industrial load identification, and least unified residue (LUR) is about 10−16, which is much better than the factorial hidden Markov model (FHMM) and the artificial neural network (ANN) model.

Revisiting Sequential Information Bottleneck: New Implementation and Evaluation.

We introduce a modern, optimized, and publicly available implementation of the sequential Information Bottleneck clustering algorithm, which strikes a highly competitive balance between clustering quality and speed. We describe a set of optimizations that make the algorithm computation more efficient, particularly for the common case of sparse data representation. The results are substantiated by an extensive evaluation that compares the algorithm to commonly used alternatives, focusing on the practically important use case of text clustering. The evaluation covers a range of publicly available benchmark datasets and a set of clustering setups employing modern word and sentence embeddings obtained by state-of-the-art neural models. The results show that in spite of using the more basic Term-Frequency representation, the proposed implementation provides a highly attractive trade-off between quality and speed that outperforms the alternatives considered. This new release facilitates the use of the algorithm in real-world applications of text clustering.

Comprehensive multivariate evaluation of the effects on cell phenotypes in multicolor flow cytometry data using ANOVA simultaneous component analysis

AbstractThis work proposes an approach to assess the effects observed in multicolor flow cytometry (MFC) experiments, for all markers and experimental factors simultaneously. It achieves this end by extending ANOVA simultaneous component analysis (ASCA), a multivariate version of ANOVA, to flow cytometry data. It is based on an initial multiset PCA model to describe the main variation patterns of cell marker expression, followed by an ASCA model on the histograms built from these PCA scores. This approach allows for determining the variations in cell phenotype distribution that are related to the experimental design. On a data set from a study of the immune response to prolonged physical exercise, the proposed method computed the effect size and statistical significance of all the experimental factors and their interactions. Most notably, it provided easily interpretable submodels for the overall effect of the walking exercise and for the interaction between exercise and the responsiveness to a bacterial stimulus. The application of a time‐guided sequential clustering algorithm to the ASCA scores revealed a stratification of the studied individuals based on their neutrophil activation dynamics. These effects were not clearly detectable using PCA alone. In comparison with pairwise classification models by DAMACY (a discriminant analysis method for MFC data), ASCA results were less detailed in describing differences between specific samples, but had the advantage of modeling several factors and levels simultaneously. Such characteristics make the proposed implementation of ASCA an effective and complementary addition to the chemometric methodologies for the analysis of MFC data.

GPSeqClus: An R package for sequential clustering of animal location data for model building, model application and field site investigations

Abstract Evaluating the concentration of animal locations, in space and time, provides insight regarding animal activity and behaviour. Most commonly collected via global positioning system technology, these concentrations are identified using rule sets resulting in location ‘clusters’. Cluster algorithms provide a framework for modelling and predicting behavioural states when paired with field data to evaluate habitat characteristics and validate results. We provide a sequential clustering algorithm package (GPSeqClus) to process location datasets based on user‐defined parameters. GPSeqClus also calculates an array of movement attributes commonly applied as covariates to develop cluster‐based models. Our package maps locations and clusters, with additional functions to assess specific clusters for site investigations and to export GPS Exchange format (.gpx) files for navigation. We highlight the ability of GPSeqClus to modify existing cluster attributes or to append additional attributes, and the flexibility to accommodate archived or near real‐time (satellite‐driven) datasets. Although spatio‐temporal clustering is widely used in animal ecology, clustering routines are commonly applied to assess carnivore movement patterns to predict denning or feeding locations, evaluate prey composition and estimate kill rates and handling times. We demonstrate the applicability of GPSeqClus by constructing clusters using mountain lion Puma concolor location data. Our package provides an efficient data processing routine to build, characterize, visualize and navigate to clusters. Attributes provided within GPSeqClus can be used when developing predictive cluster models, or can be applied to other modelling procedures to advance understandings of animal behaviour.

Statistical hierarchical clustering algorithm for outlier detection in evolving data streams

Anomaly detection is a hard data analysis process that requires constant creation and improvement of data analysis algorithms. Using traditional clustering algorithms to analyse data streams is impossible due to processing power and memory issues. To solve this, the traditional clustering algorithm complexity needed to be reduced, which led to the creation of sequential clustering algorithms. The usual approach is two-phase clustering, which uses online phase to relax data details and complexity, and offline phase to cluster concepts created in the online phase. Detecting anomalies in a data stream is usually solved in the online phase, as it requires unreduced data. Contrarily, producing good macro-clustering is done in the offline phase, which is the reason why two-phase clustering algorithms have difficulty being equally good in anomaly detection and macro-clustering. In this paper, we propose a statistical hierarchical clustering algorithm equally suitable for both detecting anomalies and macro-clustering. The proposed algorithm is single-phased and uses statistical inference on the input data stream, resulting in statistical distributions that are constantly updated. This makes the classification adaptable, allowing agglomeration of outliers into clusters, tracking population evolution, and to be used without knowing the expected number of clusters and outliers. The proposed algorithm was tested against typical clustering algorithms, including two-phase algorithms suitable for data stream analysis. A number of typical test cases were selected, to show the universality and qualities of the proposed clustering algorithm.

An analysis of MapReduce efficiency in document clustering using parallel K-means algorithm

One of the significant data mining techniques is clustering. Due to expansion and digitalization of each field, large datasets are being generated rapidly. Such large dataset clustering is a challenge for traditional sequential clustering algorithms due to huge processing time. Distributed parallel architectures and algorithms are thus helpful to achieve performance and scalability requirement of clustering large datasets. In this study, we design and experiment a parallel k-means algorithm using MapReduce programming model and compared the result with sequential k-means for clustering varying size of document dataset. The result demonstrates that proposed k-means obtains higher performance and outperformed sequential k-means while clustering documents.

A visual navigation algorithm for paddy field weeding robot based on image understanding

Navigation system and its navigation algorithm are the crucial parts for intelligent paddy field weeding robot. The environments of paddy fields are complicated in South China. The colors of weed, duckweed and cyanobacteria, which grow in paddy fields, are very similar with rice seedlings. Moreover, the rice seedlings present various morphological features during the growth progress. Therefore, how to extract the guidance lines for navigation system and weeding robot presents various challenges. In order to deal with the above mentioned problems, a navigation method for weeding robot based on SUSAN (smallest univalue segment assimilating nucleus) corner and improved sequential clustering algorithm is proposed in this paper. Firstly, gray feature in paddy field image is extracted by using the adaptive graying algorithm. Secondly, the SUSAN corners are extracted as characteristic points. Thirdly, the seedling navigation line is detected by applying the improved sequential clustering algorithm and Hough Transform. Finally, the position error and angle error are calculated, and a fuzzy controller is designed to control the robot. Experimental results show desirable performances of the proposed method. The proposed segmentation method is effective in complicated environment.

Strategies for the quality assessment of the health care service providers in the treatment of Gastric Cancer in Colombia

BackgroundWhile, at its inception in 1993, the health care system in Colombia was publicized as a paradigm to be copied across the developing world, numerous problems in its implementation have led to, what is now, an inefficient and crisis-ridden health system. Furthermore, as a result of inappropriate tools to measure the quality of the health service providers, several corruption scandals have arisen in the country. This study attempts to tackle this situation by proposing a strategy for the quality assessment of the health service providers (Entidades Promotoras de Salud, EPS) in the Colombian health system. In particular, as a case study, the quality of the treatment of stomach cancer is analyzed.MethodsThe study uses two complementary techniques to address the problem. These techniques are applied based on data of the treatment of gastric cancer collected on a nation-wide scale by the Colombian Ministry of Health and Welfare. First, Data Envelopment Analysis (DEA) and the Malmquist Index (MI) are used to establish the most efficient EPS’s within the system, according to indicators such as opportunity indicators. Second, sequential clustering algorithm, related to process mining a field of data mining, is used to determine the medical history of all patients and to construct typical care pathways of the patients belonging to efficient and inefficient EPS’s. Lastly, efforts are made to identify traits and differences between efficient and inefficient EPS’s.ResultsEfficient and inefficient EPS were identified for the years 2010 and 2011. Additionally, a Malmquist Index was used to calculate the relative changes in the efficiency of the health providers. Using these efficiency rates, the typical treatment path of patients with gastric cancer was found for two EPSs: one efficient and another inefficient. Finally, the typical traits of the care pathways were established.ConclusionsCombining DEA and process mining proved to be a powerful approach understanding the problem and gaining valuable insight into the inner workings of the Colombian Health System, especially in terms of the treatment process performed by health care providers in critical illnesses such as cancer. However, no sufficiently compelling results were found to establish the contribution of such a combination to evaluate the quality in the delivery of health services.

Segmentation-free optical character recognition for printed Urdu text

This paper presents a segmentation-free optical character recognition system for printed Urdu Nastaliq font using ligatures as units of recognition. The proposed technique relies on statistical features and employs Hidden Markov Models for classification. A total of 1525 unique high-frequency Urdu ligatures from the standard Urdu Printed Text Images (UPTI) database are considered in our study. Ligatures extracted from text lines are first split into primary (main body) and secondary (dots and diacritics) ligatures and multiple instances of the same ligature are grouped into clusters using a sequential clustering algorithm. Hidden Markov Models are trained separately for each ligature using the examples in the respective cluster by sliding right-to-left the overlapped windows and extracting a set of statistical features. Given the query text, the primary and secondary ligatures are separately recognized and later associated together using a set of heuristics to recognize the complete ligature. The system evaluated on the standard UPTI Urdu database reported a ligature recognition rate of 92% on more than 6000 query ligatures.

A Segmentation Method for Lung Parenchyma Image Sequences Based on Superpixels and a Self-Generating Neural Forest.

BackgroundLung parenchyma segmentation is often performed as an important pre-processing step in the computer-aided diagnosis of lung nodules based on CT image sequences. However, existing lung parenchyma image segmentation methods cannot fully segment all lung parenchyma images and have a slow processing speed, particularly for images in the top and bottom of the lung and the images that contain lung nodules.MethodOur proposed method first uses the position of the lung parenchyma image features to obtain lung parenchyma ROI image sequences. A gradient and sequential linear iterative clustering algorithm (GSLIC) for sequence image segmentation is then proposed to segment the ROI image sequences and obtain superpixel samples. The SGNF, which is optimized by a genetic algorithm (GA), is then utilized for superpixel clustering. Finally, the grey and geometric features of the superpixel samples are used to identify and segment all of the lung parenchyma image sequences.ResultsOur proposed method achieves higher segmentation precision and greater accuracy in less time. It has an average processing time of 42.21 seconds for each dataset and an average volume pixel overlap ratio of 92.22 ± 4.02% for four types of lung parenchyma image sequences.

A Generic Data-driven Sequential Clustering Algorithm Determining Activity Skeletons

Many activity-based models start by scheduling inflexible or mandatory activities (if present), before more flexible activities. Often work and educational activities are assumed as most stringent and recognized as the only mandatory activities. According to this definition, only 45% of all schedules contains a mandatory activity (OVG single-day travel survey in Flanders, Belgium). This means 55% of schedules does not have a traditional mandatory-flexible activity structure. This research proposes a completely data-driven approach to reveal the real basic structure of individuals’ schedules, i.e. the skeleton schedule sequence. To this end, a sequential clustering algorithm was developed. Furthermore, an in-depth analysis of the parameter settings was performed. The proposed method reveals a set of skeleton activity schedules and confirms the importance of work and education.

GPU Accelerated K Means Clustering Refined using ANT Colony Optimization

Discovery of clusters using parallelization is the answer for handling computational issues invoked by large datasets in cluster analysis. The Graphics Processing Unit and Compute Unified Device Architecture form a joint platform to improvise parallel clustering algorithms. The parallel K-Means clustering algorithm improves the speedup of clustering but often faces with problems of falling into local optima. The meta-heuristic algorithm used to optimize the clustering process using pheromone updation is known as Ant Colony Optimization technique. The K-Means clustering solution is fine tuned by applying parallel implementation of Ant Colony Optimization K-Means Clustering in order to increase efficiency. The aim is to combine optimization characteristic of Ant Colony Optimization for calculation of centroids with clustering attitude of K-Means and enhance the solution using processing power of GPU. The parallelization strategy utilized exhibits variations in speedup and quality of the clusters. The Parallel Ant Colony Clustering algorithm is tested on standard datasets and performance is compared with Sequential K-Means, Parallel K-Means, Sequential Tabu K-Means, Parallel Tabu K-Means and Sequential Ant Colony Clustering algorithms. The experimental results confirm yet another parallelization technique to unravel data clustering problems.

Pseudo-Anonymization of Social Networksby Sequential Clustering and Classification

In these days every one placing the personal data like photos, bio data etc. in so many social sites like face book, Gmail, matrimony etc. but the data is not safe in all conditions this problem is called privacy preserving problem. It provides an anonymized view of the data through a unified network, without revealing information to any of the users. Finally we develop an algorithms which is based on sequential clustering that provides centralized settings. This algorithm works based on the SaNGreeA algorithm, because Campan and Truta which is the leading algorithm for achieving anonymity in networks by means of clustering. The disadvantage of SaNGreeA, it builds clusters gradually. It cannot use actual Information Loss. This information loss will be evaluated only when all of the clusters are defined. So it contains structural information loss. But in Sequential clustering algorithm overcome this problem. In this algorithm makes decisions based on the measure of real information loss. Finally it gives a framework to classify the data with less information loss.

Jet shape and fragmentation function measurements with CMS

We present measurements of the jet fragmentation functions and transverse momentum profiles (shapes) in PbPb collisions at s NN = 2.76 TeV per nucleon-pair center-of-mass energy with the CMS detector. Jets are reconstructed by the anti- k T sequential clustering algorithm with a resolution parameter of 0.3, using particle-flow objects that combine tracking and calorimetric information. In this proceedings the jet properties of inclusive jets with transverse momentum p T > 100 GeV / c in PbPb collisions are measured for reconstructed charged particles with p T > 1 GeV / c in a cone of radius Δ ϕ 2 + Δ η 2 = 0.3 around the jet axis. A data sample of PbPb collisions collected in 2011 at a center-of-mass energy of s NN = 2.76 TeV corresponding to an integrated luminosity of 150 μb −1 is used. The results for PbPb collisions as a function of collision centrality are compared to reference distributions based on the full 2013 pp statistics collected at the same collision energy. Centrality-dependent modifications of the fragmentation functions and shapes are revealed.

Minimum profile Hellinger distance estimation for a semiparametric mixture model

AbstractIn this paper, we propose a new effective estimator for a class of semiparametric mixture models where one component has known distribution with possibly unknown parameters while the other component density and the mixing proportion are unknown. Such semiparametric mixture models have been often used in multiple hypothesis testing and the sequential clustering algorithm. The proposed estimator is based on the minimum profile Hellinger distance (MPHD), and its theoretical properties are investigated. In addition, we use simulation studies to illustrate the finite sample performance of the MPHD estimator and compare it with some other existing approaches. The empirical studies demonstrate that the new method outperforms existing estimators when data are generated under contamination and works comparably to existing estimators when data are not contaminated. Applications to two real data sets are also provided to illustrate the effectiveness of the new methodology. The Canadian Journal of Statistics 42: 246–267; 2014 © 2014 Statistical Society of Canada

A closer look into sequential clustering algorithms and associated post-processing refinement strategies

Clustering refers to a group of unsupervised classification techniques which, generally, only rely on the available data patterns to infer groups of similar patterns. Among the many available clustering algorithms the so called sequential clustering algorithms have been characterised as fast and straightforward methods which produce, as result, a single clustering. They have the drawback of being dependent on the order in which data patterns are input to the algorithm and, generally, produce compact and spherical clusters. The focus of the work is the empirical investigation of a group of three sequential algorithms namely the basic sequential algorithmic scheme (BSAS) and two of its variations, the MBSAS and the TTSAS. The work also investigates the use of three refinement strategies which aim to improve the performance of the three sequential algorithms considered, which are based on two procedures: merge and reassignment. Results from experiments conducted in various data domains (from UCI and synthetic) are presented and a comparative analysis is given as evidence of the benefits of sequential clustering algorithm coupled with a refinement procedure.