Closest Centroid Research Articles

Применение машинного обучения для анализа технических характеристик и параметров эксплуатируемого подвижного состава на высокоскоростных железнодорожных магистралях

Purpose: Machine learning is one of the most popular technologies at the present time, which finds application in various industries. One such industry is rail transport, where machine learning can significantly improve the management of locomotives and electric trains. The article is devoted to the study of the possibility of using machine learning to analyze the technical characteristics and parameters of the rolling stock using electric trains operating on high-speed railway lines as an example. The article discusses machine learning mechanisms for data analysis, as well as offers practical recommendations for using machine learning for data analysis on railway transport. Methods: The k-means method is one of the most popular machine learning algorithms for clustering, which allows to split a dataset into k similar groups or clusters. The algorithm is based on finding the centroids (mean values) of each cluster and assigning objects to the cluster that has the closest centroid. Depending on the choice of the initial values of the centroids and the parameters of the algorithm, the partitioning of data into clusters can be different. In addition, k-means is a relatively easy-to-implement and efficient clustering algorithm that can be used as an initial estimate for more complex clustering algorithms; The linear regression method is a statistical algorithm used to determine the relationship between two continuous variables. The algorithm uses a linear function that describes the relationship between the dependent variable and one or more independent variables. Results: Algorithm for processing and primary analysis of the characteristics and parameters of high-speed rolling stock, the results of parameter clustering, a technique for predicting the technical characteristics of prospective high-speed rolling stock. Practical significance: The methods of application of machine learning for the analysis of technical characteristics and parameters of the operated rolling stock on high-speed railway lines are shown. The methods will allow to substantiate and predict the necessary characteristics when developing technical requirements for high-speed transport in Russia. Using machine learning, it is possible to improve the accuracy of calculating the technical and economic models of high-speed highways, and further reduce the cost of operating rolling stock.

Comparison of the Performance of the K-Means and K-Medoids Algorithms in Grouping Regencies/Cities in Sumatera Based on Poverty Indicators

K-Means is a non-hierarchical approach that separates data into a number of groups according on how far an object is from the closest centroid. K-Medoids is a non-hierarchical clustering technique that separates data into a number of groups according on how far away an object is from the closest medoid. The two approaches were put to the test using data on poverty in Sumatra in 2021, when the Covid-19 outbreak had caused the poverty rate to increase from the year before. This research is an applied research which begins by studying relevant theories. The data used in this study is secondary data sources from the BPS website regarding poverty indicators. This study aims to determine regional groups and compare the results of grouping with the k-means and k-medoids methods. To find out the best performance between the two methods, that is by looking at the lowest Davies Bouldin Index (DBI). The results of this study are the k-means algorithm produces as many as 34 districts/cities incorporated in cluster 1, 52 districts/cities in cluster 2, 23 districts/cities in cluster 3, and 45 districts/cities in cluster 4. k-medoids, namely in clusters 1, 2, 3, and 4, respectively, as many as 53, 40, 37, and 24 districts/cities. Based on the results of the grouping, the DBI k-means of 1,584 and k-medoids of 2,359 were obtained. This means that the k-means algorithm is better than the k-medoids, because the k-means DBI is smaller than the k-medoids.

Trajectory-optimized cluster-based network model for the sphere wake

We propose a novel trajectory-optimized cluster-based network model (tCNM) for nonlinear model order reduction from time-resolved data following Li et al. [“Cluster-based network model,” J. Fluid Mech. 906, A21 (2021)] and improving the accuracy for a given number of centroids. The starting point is k-means++ clustering, which minimizes the representation error of the snapshots by their closest centroids. The dynamics is presented by “flights” between the centroids. The proposed trajectory-optimized clustering aims to reduce the kinematic representation error further by shifting the centroids closer to the snapshot trajectory and refining state propagation with trajectory support points. Thus, curved trajectories are better resolved. The resulting tCNM is demonstrated for the sphere wake for three flow regimes, including the periodic, quasi-periodic, and chaotic dynamics. The representation error of tCNM is five times smaller as compared to the approximation by the closest centroid. Thus, the error is at the same level as proper orthogonal decomposition (POD) of same order. Yet, tCNM has distinct advantages over POD modeling: it is human interpretable by representing dynamics by a handful of coherent structures and their transitions; it shows robust dynamics by design, i.e., stable long-time behavior; and its development is fully automatable, i.e., it does not require tunable auxiliary closure and other models.

Comprehensive Review of K-Means Clustering Algorithms

This paper presents a comprehensive review of existing techniques of k-means clustering algorithms made at various times. The k-means algorithm is aimed at partitioning objects or points to be analyzed into well separated clusters. There are different algorithms for k-means clustering of objects such as traditional k-means algorithm, standard k-means algorithm, basic k-means algorithm and the conventional k-means algorithm, this are perhaps the most widely used versions of the k-means algorithms. These algorithms uses the Euclidean distance as its metric and minimum distance rule approach by assigning each data points (objects) to its closest centroids.

Machine learning applied in natural language processing

Machine learning applied in natural language processing

Predictive intelligence of reliable analytics in distributed computing environments

Lack of knowledge in the underlying data distribution in distributed large-scale data can be an obstacle when issuing analytics & predictive modelling queries. Analysts find themselves having a hard time finding analytics/exploration queries that satisfy their needs. In this paper, we study how exploration query results can be predicted in order to avoid the execution of ‘bad’/non-informative queries that waste network, storage, financial resources, and time in a distributed computing environment. The proposed methodology involves clustering of a training set of exploration queries along with the cardinality of the results (score) they retrieved and then using query-centroid representatives to proceed with predictions. After the training phase, we propose a novel refinement process to increase the reliability of predicting the score of new unseen queries based on the refined query representatives. Comprehensive experimentation with real datasets shows that more reliable predictions are acquired after the proposed refinement method, which increases the reliability of the closest centroid and improves predictability under the right circumstances.

Embedding hierarchical clustering in product quantization for feature indexing

Effective indexing is a crucial need for large scale object matching and retrieval. In this work, a novel indexing scheme is presented, that exploits the advantages of hierarchical clustering and product quantization. First, the high dimensional feature space is decomposed into disjointed sub-spaces and the data belonging to each sub-space is separately represented by a hierarchical clustering tree. Second, each tree quantizes a distinct part of an input vector to the closest centroid of a leaf node and the distances for all the pairs of centroids are pre-computed and stored in a lookup table. Finally, searching for a given query is proceeded in parallel between the trees and is performed efficiently in the quantized space using the pre-computed lookup tables. The proposed method has been validated by a number of experiments, demonstrating significant improvements of search performance in comparison with other methods.

A Cluster-Based Boosting Algorithm for Bankruptcy Prediction in a Highly Imbalanced Dataset

Bankruptcy prediction has been a popular and challenging research topic in both computer science and economics due to its importance to financial institutions, fund managers, lenders, governments, as well as economic stakeholders in recent years. In a bankruptcy dataset, the problem of class imbalance, in which the number of bankruptcy companies is smaller than the number of normal companies, leads to a standard classification algorithm that does not work well. Therefore, this study proposes a cluster-based boosting algorithm as well as a robust framework using the CBoost algorithm and Instance Hardness Threshold (RFCI) for effective bankruptcy prediction of a financial dataset. This framework first resamples the imbalance dataset by the undersampling method using Instance Hardness Threshold (IHT), which is used to remove the noise instances having large IHT value in the majority class. Then, this study proposes a Cluster-based Boosting algorithm, namely CBoost, for dealing with the class imbalance. In this algorithm, the majority class will be clustered into a number of clusters. The distance from each sample to its closest centroid will be used to initialize its weight. This algorithm will perform several iterations for finding weak classifiers and combining them to create a strong classifier. The resample set resulting from the previous module, will be used to train CBoost, which will be used to predict bankruptcy for the validation set. The proposed framework is verified by the Korean bankruptcy dataset (KBD), which has a very small balancing ratio in both the training and the testing phases. The experimental results of this research show that the proposed framework achieves 86.8% in AUC (area under the ROC curve) and outperforms several methods for dealing with the imbalanced data problem for bankruptcy prediction such as GMBoost algorithm, the oversampling-based method using SMOTEENN, and the clustering-based undersampling method for bankruptcy prediction in the experimental dataset.

Are triple-negative tumours and basal-like breast cancer synonymous? Authors' response

We read with interest the issues raised by Rakha and colleagues [1] in their response to our recent research article [2], and we are pleased to address them. An important conclusion from our research article is that triple-negative breast cancer can be equated to breast cancer. In their letter, Rakha and colleagues [1] state that equating triple-negative phenotype (TNP) tumours with breast cancer is misleading and is not supported by the data we have presented. It is important to realize that, as we have also pointed out in our article [2], the breast cancer subtype was initially defined based on the gene expression pattern of the so-called 'intrinsic gene list' in only six breast tumours [3]. Since this initial report, the intrinsic gene list that is used to identify breast tumours has been updated multiple times [3-5]. This shows that a gene-expression-based definition of breast cancer has its limitations. There are two reasons for Rakha and colleagues [1] to dispute our results. First, there are nine triple-negative tumours that have a gene expression pattern that is not strongly correlated to the centroid. However, when we do not set the threshold for the correlation coefficient to the molecular subtype centroids at 0.1, but simply look at the closest centroid, all our unclassifiable samples will be allocated to the subtype. This leaves only the four samples that are allocated to the normal epithelial-like subtype. This is a problematic group, as it was originally defined on the basis of samples that did not contain tumour cells after neoadjuvant chemotherapy treatment and there are many similarities between this subtype and the subtype [3]. Second, Rakha and colleagues [1] state that '18.6% of non-TNP cancers cluster together with TNP cancers in the basal-like cluster'. This argument is based on an incorrect understanding of our findings. These 18.6% are not part of a series of tumours that were all negative for a TNP, but 18.6% of a series of tumours not selected for TNP. These tumours include both TNP and non-TNP samples. In Additional file 1 [2], it is shown that some of these tumours have the gene expression pattern of triple-negative tumours – these are in fact the TNP tumours from this unselected series of tumours. This lends even greater support to the notion that triple-negative tumours are synonymous with tumours. We therefore stand by our results and believe that gene expression profiling studies have highlighted the importance of basal-like/TNP tumours but that simple immunohisto-chemistry is sufficient to classify these tumours.

An efficient k-means clustering algorithm using simple partitioning

The k-means algorithm is one of the most widely used methods to partition a dataset into groups of patterns. However, most k-means methods require expensive distance calculations of centroids to achieve convergence. In this paper, we present an efficient algorithm to implement a k-means clustering that produces clusters comparable to slower methods. In our algorithm, we partition the original dataset into blocks; each block unit, called a unit block (UB), contains at least one pattern. We can locate the centroid of a unit block (CUB) by using a simple calculation. All the computed CUBs form a reduced dataset that represents the original dataset. The reduced dataset is then used to compute the final centroid of the original dataset. We only need to examine each UB on the boundary of candidate clusters to find the closest final centroid for every pattern in the UB. In this way, we can dramatically reduce the time for calculating final converged centroids. In our experiments, this algorithm produces comparable clustering results as other k-means algorithms, but with much better performance.