Initial Number Of Clusters Research Articles

A Hierarchical Gamma Mixture Model-Based Method for Classification of High-Dimensional Data

Data classification is an important research topic in the field of data mining. With the rapid development in social media sites and IoT devices, data have grown tremendously in volume and complexity, which has resulted in a lot of large and complex high-dimensional data. Classifying such high-dimensional complex data with a large number of classes has been a great challenge for current state-of-the-art methods. This paper presents a novel, hierarchical, gamma mixture model-based unsupervised method for classifying high-dimensional data with a large number of classes. In this method, we first partition the features of the dataset into feature strata by using k-means. Then, a set of subspace data sets is generated from the feature strata by using the stratified subspace sampling method. After that, the GMM Tree algorithm is used to identify the number of clusters and initial clusters in each subspace dataset and passing these initial cluster centers to k-means to generate base subspace clustering results. Then, the subspace clustering result is integrated into an object cluster association (OCA) matrix by using the link-based method. The ensemble clustering result is generated from the OCA matrix by the k-means algorithm with the number of clusters identified by the GMM Tree algorithm. After producing the ensemble clustering result, the dominant class label is assigned to each cluster after computing the purity. A classification is made on the object by computing the distance between the new object and the center of each cluster in the classifier, and the class label of the cluster is assigned to the new object which has the shortest distance. A series of experiments were conducted on twelve synthetic and eight real-world data sets, with different numbers of classes, features, and objects. The experimental results have shown that the new method outperforms other state-of-the-art techniques to classify data in most of the data sets.

Analysis of Building Electricity Use Pattern Using K-Means Clustering Algorithm by Determination of Better Initial Centroids and Number of Clusters

Energy demands in the building sector account for more than 30% of the total energy use and more than 55% of the global electricity demand. Efforts to develop sustainable buildings are progressing but are still not keeping pace with the growing building sector and the rising demand for energy. Analyzing the energy use pattern of buildings and planning for energy conservation in existing buildings are essential. In this research, we propose a method to analyze the energy use pattern in a building using the K-means clustering method. Initial centroids in K-means clustering are chosen randomly so that the clustering result changes every time. This instability is removed in the proposed method by the selection of initial centroids using a percentile method based on empirical cumulative distribution. The results from the proposed method have better accuracy, and the internal cohesion and separation between clusters are better than the random initialization method. Analyzing yearly electricity use using the proposed clustering method, the daily pattern of electricity use can be categorized according to the operation of buildings. For this purpose, in this research, electricity use pattern was analyzed for three to six clusters. In comparison with the university schedule, six clusters were found to be appropriate and the accuracy was 89.3%. Once daily electricity use are categorized, base electricity consumption, electricity consumption by human activities, and electricity consumption by air-conditioning can be determined. As energy consumption by usage is clarified, measures for energy consumption in university buildings can be proposed.

Defining Geographical Rating Territories in Auto Insurance Regulation by Spatially Constrained Clustering

Territory design and analysis using geographical loss cost are a key aspect in auto insurance rate regulation. The major objective of this work is to study the design of geographical rating territories by maximizing the within-group homogeneity, as well as maximizing the among-group heterogeneity from statistical perspectives, while maximizing the actuarial equity of pure premium, as required by insurance regulation. To achieve this goal, the spatially-constrained clustering of industry level loss cost was investigated. Within this study, in order to meet the contiguity, which is a legal requirement on the design of geographical rating territories, a clustering approach based on Delaunay triangulation is proposed. Furthermore, an entropy-based approach was introduced to quantify the homogeneity of clusters, while both the elbow method and the gap statistic are used to determine the initial number of clusters. This study illustrated the usefulness of the spatially-constrained clustering approach in defining geographical rating territories for insurance rate regulation purposes. The significance of this work is to provide a new solution for better designing geographical rating territories. The proposed method can be useful for other demographical data analysis because of the similar nature of the spatial constraint.

Retraction: Xiaohong Zhu, Zili Yang and Tabharit Zoubir, Research on the matching algorithm for heterologous image after deformation in the same scene

The existing contour matching algorithm is difficult to deal with the local contour matching problem of the heterologous image in same scene, and the different shooting angles or non approximate transformation will cause some deformation. In this regard, this paper proposes a matching algorithm based on image segmentation. According to the characteristics of the image, the idea of extracting the outline by first image segmentation and then extracting the contour is used to extract the coarse contour. the feature space based on mean gray, gray variance and entropy is constructed, to represent the characteristics of different material object in the image; then the initial clustering number and clustering center optimized by the ant colony algorithm is used to make fuzzy clustering of the image feature space; Canny operator is used for the edge detection, so the coarse contour images is obtained. The PNP algorithm is used to combine the same name points obtained from the initial matching to calculate the angle transformation parameters between the two images. By inverse operation, the matching contour in the two images is corrected to the same view. The experimental results show that the proposed method can effectively improve the precision of the segmentation and the matching precision of the heterogenous image.

Natural fruitlet abscission as related to apple tree carbon balance estimated with the MaluSim model

Apple trees produce many more flower clusters than needed for a full crop, but natural early season flower and fruitlet abscission drastically reduce the final fruit number. Natural fruit abscission varies significantly year to year. There have been attempts to try to model apple fruit abscission in the past. However, due to the great complexity of a perennial crop system in a dynamic environment with significant plant manipulations, regulatory processes and controlling environmental variables have been difficult to elucidate. In 1995, a field trial was planted at the New York State Agricultural Experiment Station in Geneva, New York with 3 apple cultivars (‘Delicious’, ‘Gala’, and ‘McIntosh’). Beginning in 2000 and for 18 years thereafter, we recorded the natural whole-season fruit abscission of untreated trees that received no chemical or hand thinning. We also estimated early season patterns of carbohydrate supply-to-demand each year with a carbon balance model. These data were used to correlate tree carbon balance status and other environmental variables with natural fruit abscission responses. In general terms, natural set, defined as final fruit/flower cluster, of ‘Gala’ averaged ˜1 fruit for each flower cluster (fruit set = 0.9), whereas fewer fruits were set on ‘Delicious’ and ‘McIntosh’ (fruit set = 0.7 and 0.6, respectively). Fruit set of ‘Gala’ was less variable than of ‘Delicious’ or ‘McIntosh’, and there was a clear pattern for decreasing fruit set when the number of initial flower clusters per tree increased. Fruit weight was less dependent on fruit number for ‘Delicious’ and ‘McIntosh’ than for ‘Gala’. Multiple regression models indicated that number of flower clusters per tree and average carbohydrate balance between 0–60 degree days (DD) after bloom and 300–360 DD after bloom were the main significant variables that explained 60–80% of the variability in natural fruit set or final fruit number. For ‘Delicious’, temperatures of the previous fall also explained a significant amount of variation in final fruit set and final fruit number. For ‘Gala’, carbon balance from bloom to shortly after petal fall and when fruits were about 18 mm were the two main periods, which were more sensible to carbohydrate deficiency triggering fruit abscission. A later susceptible period was also observed for ‘McIntosh’, suggesting a larger thinning window for this cultivar.

Clustering of human activities from emerging movements

This paper is positioned in the well-established field of smart home. This area of research, highly multidisciplinary, has raised a lot of attention from researchers due to the broad real life application it could serve. One of them is the assistance of the cognitively impaired persons such as head trauma victims or persons afflicted by dementia (e.g.: Alzheimer’s disease). However, to propose powerful technological cognitive orthoses, the decades old challenge of human activity recognition must be addressed. In this paper, we propose a clustering method exploiting the Flocking algorithm for Activity of Daily Living (ADL) learning and recognition. In particular, our new method enables to both exploit events based data from the multimodal sensors of a smart home and qualitative spatial information extracted from a tracking method. Among the advantages of the method, the Flocking based algorithm does not require an initial number of clusters, unlike other clustering algorithms such as K-means. Two sets of real case scenarios were collected in our smart home laboratory, the LIARA. These sets were used to compare our method with traditional unsupervised algorithms and to evaluate the usefulness of the qualitative spatial information. The study shows that for traditional event based smart home data, the method outperforms the popular K-Means and Expectation-Maximization (EM) algorithms. Furthermore, the results indicate that not only the spatial data generalize, but it also further improve the performance regarding fine-grained ADLs recognition.

Screening of eco-friendly thinning agents and adjusting mechanical thinning on ‘Gala’, ‘Golden Delicious’ and ‘Fuji’ apple trees

Abstract Fruit thinning is the most important yet difficult practice that drives orchard profitability. High labor costs and difficulty to improve return bloom by hand thinning have left chemical thinning as the main method used by growers. However, unpredictability and safety/environment concerns regarding chemical thinning have set mechanical thinning as a sound alternative. Thirteen field experiments were performed during 2004–2016 in order to evaluate several agents for their use as new thinners, and adjust mechanical thinning on ‘Gala’, ‘Golden Delicious’ and ‘Fuji’. Olive oil applied at bloom reduced crop load, but russetting was also increased. Therefore, while their use is not advisable for russetting prone cultivars such as ‘Golden Delicious’, it could be a good thinner for cultivars like ‘Red Delicious’. Lime sulfur did not have a consistent thinning effect in our study when applied at bloom. Overall, no differences regarding economic value between hand, chemical, and mechanical blossom thinning were observed, suggesting mechanical thinning as a valid alternative approach. For ‘Gala’ strains, 6 km h−1 and 250 rpm with 270 strings was the best configuration to provide an ideal crop load of ∼6 fruit/cm2 of TCSA and an average fruit size of 170 g. For ‘Fuji’, 5 km h−1 and 320 rpm with 270 strings provided a crop load in accordance to the optimum range for this cultivar in our conditions. However, combination of mechanical thinning plus chemical treatments might be the ideal strategy for ‘Fuji’ strains when the initial number of flower clusters per tree is above 500. For ‘Golden Delicious’ strains, 6 km h−1 and 230 rpm with 270 strings was the best configuration to provide an ideal crop load within the optimum range. Mechanical thinning timing was also examined at different phenological stages (E2, F1, F2, and G), with no significant differences regarding yield, fruit size or crop load between them. Two prediction models (‘Gala’ & ‘Golden Delicious’) were developed to adjust the right tractor and rotational speeds depending on the initial number of flower clusters. The method begins with first calculating the final fruit number needed per tree (crop load for each particular cultivar) in order to achieve the desired yield. Then, tractor and rotational speeds can be determined by the model once knowing the initial number of flower clusters per tree.

Cluster analysis of polymers using laser-induced breakdown spectroscopy with K-means

Laser-induced breakdown spectroscopy (LIBS) combined with K-means algorithm was employed to automatically differentiate industrial polymers under atmospheric conditions. The unsupervised learning algorithm K-means were utilized for the clustering of LIBS dataset measured from twenty kinds of industrial polymers. To prevent the interference from metallic elements, three atomic emission lines (C i 247.86 nm , H i 656.3 nm, and O i 777.3 nm) and one molecular line C–N (0, 0) 388.3 nm were used. The cluster analysis results were obtained through an iterative process. The Davies–Bouldin index was employed to determine the initial number of clusters. The average relative standard deviation values of characteristic spectral lines were used as the iterative criterion. With the proposed approach, the classification accuracy for twenty kinds of industrial polymers achieved 99.6%. The results demonstrated that this approach has great potential for industrial polymers recycling by LIBS.

A novel data clustering algorithm using heuristic rules based on [formula omitted]-nearest neighbors chain

In practice, clustering algorithms usually suffer from the complex structure of the dataset, including data distribution and dimensionality. Meanwhile, the number of clusters, which is required as an input, is usually unavailable. In this paper, we propose a novel data clustering algorithm: it uses heuristic rules based on k-nearest neighbors chain and does not require the number of clusters as the input parameter. Inspired by the PageRank algorithm, we first use random walk model to measure the importance of data points. Then, on the basis of the important data points, we build a K-Nearest Neighbors Chain (KNNC) to order the k nearest neighbors by distance and propose two heuristic rules to find the proper number of clusters and initial clusters. The first heuristic rule is the gap of KNNC which reflects the degree of separation of clusters with convex shapes and the second one is the nearest neighbor gap of KNNC which reflects the inner compactness of a cluster. Comprehensive comparison results on synthetic and real datasets indicate that the proposed clustering algorithm can find the proper number of clusters and achieve comparable or even better performance than the popular clustering algorithms.

An uncertainty perspective to PCM and APCM clustering

Possibilistic c-means (PCM) based clustering algorithms are widely used in the literature. Recently, adaptive PCM (APCM) is proposed to adapt the bandwidth at each iteration and the cluster merge is automatically achieved. The cluster elimination ability of APCM makes PCM very flexible to set the initial cluster number mini. However, this comes at a price of introducing another parameter α which ranges in (0,+∞). This study tries to utilize the uncertainty in the data to achieve more control over the clustering process by appropriately characterizing the uncertainty of memberships via the conditional fuzzy set. This uncertainty perspective motivates us to introduce parameters σv and α to characterize uncertainty of estimated bandwidth and noise level of the dataset respectively, which results in a unified framework of PCM and APCM (UPCM). UPCM is further developed by eliminating the σv parameter, then we get PCM clustering based on noise level (NPCM). As a result, the algorithm needs two kinds of information that is intuitive to specify for the clustering task, i.e., information of the cluster number and information of the property of clusters, and they are represented by two parameters, i.e., mini specifies the possibly over-specified cluster number, and α characterizes the closeness of clusters in the clustering result. Both parameters are not required to be exactly specified, and α ranges in [0,1]. Experiments show that the clustering process can be effectively controlled by the parameters.

An Adaptive Multiobjective Genetic Algorithm with Fuzzy c-Means for Automatic Data Clustering

This paper presents a fuzzy clustering method based on multiobjective genetic algorithm. The ADNSGA2-FCM algorithm was developed to solve the clustering problem by combining the fuzzy clustering algorithm (FCM) with the multiobjective genetic algorithm (NSGA-II) and introducing an adaptive mechanism. The algorithm does not need to give the number of clusters in advance. After the number of initial clusters and the center coordinates are given randomly, the optimal solution set is found by the multiobjective evolutionary algorithm. After determining the optimal number of clusters by majority vote method, the Jm value is continuously optimized through the combination of Canonical Genetic Algorithm and FCM, and finally the best clustering result is obtained. By using standard UCI dataset verification and comparing with existing single-objective and multiobjective clustering algorithms, the effectiveness of this method is proved.

Effective algorithm for determining the number of clusters and its application in image segmentation

The k-means algorithm is a popular clustering method for image segmentation. However, the main disadvantage of this algorithm is its dependence on the number of initial clusters. In this paper, we present an optimal criterion which can select the best segmentation result with less number of clusters. The optimal criterion overcomes the shortcoming of initialization based on the intra-class and inter-class difference. Eight digital images were employed to verify the segmentation results of the optimal criterion. Simultaneously, we have improved the traditional k-means algorithm to find the initial clustering centers efficiently. Experimental results show that the segmented images selected by the optimal criterion have sufficient stability and robustness. In addition, we verify the consistency of results by two kinds of objective assessment measures. The proposed optimal criterion can successfully display the best segmentation results precisely and efficiently so as to instead of artificial selection.

Preliminary analytical method for unsupervised remote sensing image classification based on visual perception and a force field

The analysis of remote sensing (RS) images, which is often accomplished using unsupervised image classification techniques, requires an effective method to determine an appropriate number of classification clusters. This paper proposes a preliminary analytical method to evaluate the input parameters for unsupervised RS image classification. Our approach involves first analysing the colour spaces of RS images based on the human visual perception theory. This enables the initial number of clusters and their corresponding centres to be automatically established based on the interaction of different forces in our supposed force field. The proposed approach can automatically determine the appropriate initial number of clusters and their corresponding centres for unsupervised image classification. A comparison of the experimental results with those of existing methods showed that the proposed method can considerably facilitate unsupervised image classification for acquiring accurate results efficiently and effectively without any prior knowledge.

Application of Molecular Topological Descriptors for Clustering a Database of Isothiourea Derivatives in Studying Structure – Activity Relationships

A database of some S,N,N,N′-tetrasubstituted isothiourea derivatives possessing neuroprotective properties was successfully clustered in order to study the quantitative structure – activity relationship. Clustering by k-means was carried out in the factor space of topological descriptors. The identified clusters were combined according to analyses of intra- and intercluster distances. The initial number of clusters in the k-means clustering was determined from the number of iterations for which a solution was obtained. The homogeneity of the database and the identified clusters was estimated by using a coefficient of molecular diversity. A plot of the database compounds as points in factor space led to a conclusion about the successful applicability of the proposed clustering approach.

DBSCAN BASED SEED INITIALIZATION OF K-MEANS ALGORITHM

This paper proposes effective approach to overcome the problem of finding initial number of clusters for Supervised Data mining algorithms. We present critical review of various approaches that finds the optimal number of clusters for clustering algorithms. In this paper we have used Dbscan algorithms to obtain initial seeds for basic k-means algorithm. To evaluate the proposed approache we have used iris data set, liver disorder dataset and seed dataset.

An improved method for SIFT-based copy–move forgery detection using non-maximum value suppression and optimized J-Linkage

In looking to improve the detection performance of the keypoint-based method involving smooth tampered regions, there are three problems to be addressed, namely the nonuniform distribution of the keypoints, the discriminative power of low contrast keypoints, and the high computational cost of clustering. In this study, the classical implementation framework of the keypoint-based method is improved by introducing new techniques and algorithms in order to overcome these problems. First, to acquire uniformly distributed keypoints in the test image, we propose a new solution of selecting the keypoints by region instead of contrast. To this end, we first separate the keypoint detection and selection processes. After obtaining all discernible keypoints, we adapt the non-maximum value suppression algorithm to select keypoints by combining the contrast and density of each keypoint. Second, we apply the opponent scale-invariant feature transform descriptor to enhance the discriminative power of keypoints by adding color information. Finally, to alleviate the computational cost of clustering, we optimize the J-Linkage algorithm by altering the method of computing initial clusters and affine transformation hypotheses. For this purpose, we propose the matched pair grouping algorithm that can obtain a smaller number of initial clusters by utilizing the correspondence between the superpixels in the original and duplicated regions. Experiments performed on three representative datasets confirm that the proposed method can significantly improve the detection performance in smooth tampered regions, and considerably reduce the clustering time in the case of a mass of matched pairs, compared with the state-of-the-art methods.

Использование молекулярных топологических дескрипторов для кластеризации базы данных производных изотиомочевины в рамках изучения зависимости структура — активность

Произведена успешная кластеризация базы данных производных S,N,N,N’-тетразамещенных изотиомочевин, обладающими нейропротекторными свойствами, с целью изучения количественной связи структура — активность. Кластеризация методом к-средних проводилась в факторном пространстве топологических дескрипторов с последующим объединением выделенных кластеров на основе анализа внутри- и межкластерных расстояний. Первоначальное число кластеров в методе к-средних определялось, исходя из количества итераций, за которое было получено решение. Для оценки однородности базы данных и выделенных кластеров использовался коэффициент разнообразия. Графическое представление соединений, входящих в базу данных, в виде точек факторного пространства позволяет сделать вывод об успешности данного метода.

Based on rough set and fuzzy clustering of MRI brain segmentation

A new method of MRI brain segmentation integrates fuzzy [Formula: see text]-means (FCM) clustering and rough set theory. In this paper, we use rough set algorithm to find the suitable initial clustering number to initial clustering centers for FCM. Then we use FCM to MRI brain segmentation, but the algorithm of FCM has the limitation of converging to local infinitesimal point in medical segmentation. While avoiding being trapped in a local optimum, we use the particle swarm optimization algorithm to restrict convergence of FCM which can reduce calculation. The final experiment results show that improved algorithm not only retains the advantages of rapid convergence but also can control the local convergence and improve the global search ability. The method in this paper is better than that of cluttering performance.

A Novel Cluster Head Selection Algorithm Based on Fuzzy Clustering and Particle Swarm Optimization

An important objective of wireless sensor network is to prolong the network life cycle, and topology control is of great significance for extending the network life cycle. Based on previous work, for cluster head selection in hierarchical topology control, we propose a solution based on fuzzy clustering preprocessing and particle swarm optimization. More specifically, first, fuzzy clustering algorithm is used to initial clustering for sensor nodes according to geographical locations, where a sensor node belongs to a cluster with a determined probability, and the number of initial clusters is analyzed and discussed. Furthermore, the fitness function is designed considering both the energy consumption and distance factors of wireless sensor network. Finally, the cluster head nodes in hierarchical topology are determined based on the improved particle swarm optimization. Experimental results show that, compared with traditional methods, the proposed method achieved the purpose of reducing the mortality rate of nodes and extending the network life cycle.

Identifying Cell Populations in Flow Cytometry Data Using Phenotypic Signatures.

Single-cell flow cytometry is a technology that measures the expression of several cellular markers simultaneously for a large number of cells. Identification of homogeneous cell populations, currently done by manual biaxial gating, is highly subjective and time consuming. To overcome the shortcomings of manual gating, automatic algorithms have been proposed. However, the performance of these methods highly depends on the shape of populations and the dimension of the data. In this paper, we have developed a time-efficient method that accurately identifies cellular populations. This is done based on a novel technique that estimates the initial number of clusters in high dimension and identifies the final clusters by merging clusters using their phenotypic signatures in low dimension. The proposed method is called SigClust. We have applied SigClust to four public datasets and compared it with five well known methods in the field. The results are promising and indicate higher performance and accuracy compared to similar approaches reported in literature.