Initial Number Of Clusters Research Articles

Study on cluster centers optimization of max-min distance k-means clustering algorithm based on inter-cluster separation measure

Clustering is an essential unsupervised technique when category information is not available. Although K-means and Max-min distance K-means clustering algorithms are widely used, they have some disadvantages such as dependence on the initial centers, sensitivity to outliers caused by using only distance as the clustering criterion. To overcome the problems, this paper proposes SMM-K-means algorithm which overcomes the dependence on the initial cluster centers and the initial number of clusters and the sensitivity to the outliers. First, the initial value K of the optimal cluster number is determined by the elbow method, and K-means is used for initial clustering. A new inter-cluster separation measure is then constructed based on the idea of q-nearest neighbors, which is constructed by comprehensive considering the separation between clusters and the distribution compactness of clusters themselves. Finally, the two sample points with highest degree of separation are brought into Max-min distance K-means algorithm as new initial centers for clustering. The definite determining method of cluster centers eliminates the complicated iterative calculation, and the construction of inter-cluster separation measure overcomes the sensitivity of clustering results to noise points and isolated points, and has good applicability and generalization. In addition, this algorithm is not limited by the shape and size of the clusters and has better flexibility. The experimental results show that the SMM-K-means algorithm has higher CH values, resulting in a better clustering effect and stability.

Adaptive fuzzy weighted C-mean image segmentation algorithm combining a new distance metric and prior entropy

The fuzzy C-mean clustering algorithm (FCM) is effective in image segmentation. However, its sensitivity to initialization settings and the limitation of the Euclidean distance metric in measuring similarity lead to unsatisfactory image segmentation. To overcome these shortcomings, we propose an adaptive fuzzy weighted C-mean image segmentation algorithm that combines a new distance metric and prior entropy. This algorithm significantly contributes in three aspects. First, we employ an improved sparse subspace clustering (SSC) algorithm based on the superpixel image for initialization settings. This step aims to obtain the initial number of clusters and the membership matrix, ensuring the algorithm’s convergence to the optimal solution. Second, to better capture differences between image regions, we introduce a novel distance metric that combines Euclidean and non-Euclidean distances. Simultaneously, the fuzzy weight is introduced to mitigate redundant feature interference, making the clustering result more reasonable. Finally, leveraging prior entropy – a maximum entropy under conditional constraints – we refine the algorithm’s adaptability to unknown features, thereby improving clustering accuracy. Comparative experiments with several state-of-the-art algorithms validate the effectiveness and robustness of our proposed algorithm.

Clustering-based value investing strategy in the Helsinki Stock Exchange: k-means algorithm

The purpose of this research is to study the possibility of combining quantitative clustering of stocks and value investing. The feasibility of this approach is tested using Finnish market data from the period 2005 to 2021. The benchmark index used in this research is the OMX Helsinki Growth Index. The strategy is based on the combination of P/E, P/CF, and P/B ratios, which serve as the basis for the k-means algorithm. The data is pre-processed by removing stocks that have not generated positive earnings and cash flow during the previous 12 months. The k-means algorithm assigns stocks to clusters, and the cluster with the lowest financial ratios is chosen as the value portfolio. The research also includes a sensitivity analysis of value portfolios, where the initial number of clusters in the clustering phase ranges from three to ten. Returns of different value portfolios are compared to each other and the benchmark index. The quality of results is evaluated using the Sharpe ratio and Jensen's alpha. According to the findings, the value portfolio constructed using nine clusters generated the highest risk-adjusted return, with an annual return of 30.27% over the 2005 to 2021 period. Furthermore, the best-performing value portfolio from 2005 to 2017 was compared to the benchmark index from 2018 to 2021. The value portfolio achieved an annual return of 26.05% during the 2018-2021 period, while the corresponding return of the benchmark index was 11.74%.

K-MEANS CLUSTER COUNT OPTIMIZATION WITH SILHOUETTE INDEX VALIDATION AND DAVIES BOULDIN INDEX (CASE STUDY: COVERAGE OF PREGNANT WOMEN, CHILDBIRTH, AND POSTPARTUM HEALTH SERVICES IN INDONESIA IN 2020)

One of the causes of the increasing maternal mortality rate in Indonesia is the declining performance of maternal health services in each Indonesian province. To overcome the decline in performance, namely by determining in advance the provinces that need to be prioritized for services by grouping 34 provinces in Indonesia. This study aims to obtain the best provincial grouping results so that it can prioritize the right provinces. One of the methods that are suitable for grouping provinces is K-Means because it is simple and easy to implement. The disadvantage of K-Means is that it is sensitive to determining the right number of initial clusters, so Silhouette Index and Davies Bouldin Index validation is used to obtain the optimal number of clusters with stable and consistent results. This study used healthcare data for pregnant women, childbirth, and postpartum with K=2, 3, and 4 as the initial cluster number. K-Means objects are grouped in similarities using Euclidean and Manhattan distances. The result obtained was the optimal number of clusters with K=2 using Manhattan, where the highest Silhouette Index value was 0,658685 and the lowest Davies Bouldin Index was 0,3561214 which met the criteria for determining the optimal cluster.

Research on the Siting of Rural Public Cultural Space Based on the Path-Clustering Algorithm: A Case Study of Yumin Township, Yushu City, Jilin Province, China

Cultural revitalization is the foundation of rural revitalization. Rural public cultural spaces are important places for rural residents to participate in cultural activities and important carriers for the occurrence, inheritance and development of rural culture. In this study, we propose a siting analysis method for public cultural spaces based on a path-clustering algorithm. Then, Yumin Township in Jilin Province is used as a case study to test the method. Within the township area, public cultural spaces are divided into three types based on the difference in service scope: “single-village decentralized”, “single-village concentration” and “multi-village concentration”. This paper focuses on the analysis of the siting of the “multi-village concentration” type of public cultural spaces using a path-clustering algorithm. First, the path-clustering algorithm is written, and then its parameters, initial clustering centers and number of clusters are analyzed for their effect on the clustering results. Two parameters, within-cluster sum of path distance (WCSPD) and maximum value of the path distance within-cluster (MPDWC), are proposed to evaluate the results of the algorithm. Then, taking Yumin Township of Jilin Province as an example, the villages in the township are clustered with travel distance as the main influencing factor. It is ensured that the shortest path distance from the central village in each cluster to other villages is no greater than the set travel distance. Finally, the public cultural space is constructed in the central village to serve all villages in the cluster. This siting analysis method can output a variety of results that can be used as a reference for decision making in rural public cultural space planning.

Cluster Predictive Model Using Affinity Propagation Algorithm to Group Mushroom 5.8s rRNA Sequences

Background: The main emphasis of the article is biological information on a distinct species of mushroom (Phylum Basidiomycota) data collection of 5.8s rRNA sequences. Macrofungi from the phylum Basidiomycota are predominantly used as therapeutic mushrooms in several countries. During the rainy season, hundreds of macrofungal basidiocarps were discovered in Tamilnadu. The internal transcribed spacer (ITS) and 5.8S rRNA gene sequence markers, which have been collected from NCBI, were used to isolate at least thirty of these strains that fall under the Basidiomycota kingdom (suborders of Polyporales, Hymenochataeles, and Russuales), which have the therapeutic properties of the Basidiomycota kingdom. Objectives: This article’s main objective is to organise the sequences according to similarity utilising multiple sequence alignment and an algorithmic perspective. Methods: In this paper, we use 30  30 pairwise similarity matrix data of these thirty 5.8s rRNA mushroom sequences obtained using the clustal omega tool to develop an affinity propagation approach. As a continuation of earlier work, this will be evaluated against k-means, hierarchical clustering based on the ideal cluster, and time and space complexity. Findings: The affinity propagation algorithm typically discourages providing the initial number of clusters; therefore, the optimal number of cluster values and grouping of clustered results obtained from the affinity propagation algorithm are also the same as the results obtained from the previous existing research work using the kmeans, hierarchical agglomerative clustering algorithm. Novelty: The overall suggested technique involves applying the cluster validation metrics Silhouette score, Calinski-Harabasz Index, and Davies-Bouldin Index methodologies to find the ideal number of clusters. The CD-hit Clustering tool does not offer these metrics, and the Cluster Omega tool does not support this kind of extension work. This follow-up work assists bioinformatics researchers in obtaining favourable results by utilising the existing software prior to working in wet laboratories; rather than wasting a lot of chemical resources, this result will open the door for a targeted approach. Keywords: Affinity propagation; Cluster metrics; Kmeans; mushroom sequences; Bioinformatics; Data science; Silhouette score; Calinski-Harabasz index; Davies-Bouldin index

МЕТОДЫ И АЛГОРИТМЫ КЛАСТЕРИЗАЦИИ ТЕКСТОВЫХ ДАННЫХ (ОБЗОР)

The article deals with one of the important tasks of artificial intelligence – machine processingof natural language. The solution of this problem based on cluster analysis makes it possibleto identify, formalize and integrate large amounts of linguistic expert information under conditionsof information uncertainty and weak structure of the original text resources obtained fromvarious subject areas. Cluster analysis is a powerful tool for exploratory analysis of text data,which allows for an objective classification of any objects that are characterized by a number offeatures and have hidden patterns. A review and analysis of modern modified algorithms for agglomerativeclustering CURE, ROCK, CHAMELEON, non-hierarchical clustering PAM, CLARAand the affine transformation algorithm used at various stages of text data clustering, the effectivenessof which is verified by experimental studies, is carried out. The paper substantiates therequirements for choosing the most efficient clustering method for solving the problem of increasing the efficiency of intellectual processing of linguistic expert information. Also, the paper considersmethods for visualizing clustering results for interpreting the cluster structure and dependencieson a set of text data elements and graphical means of their presentation in the form ofdendograms, scatterplots, VOS similarity diagrams, and intensity maps. To compare the quality ofthe algorithms, internal and external performance metrics were used: "V-measure", "AdjustedRand index", "Silhouette". Based on the experiments, it was found that it is necessary to use ahybrid approach, in which, for the initial selection of the number of clusters and the distribution oftheir centers, use a hierarchical approach based on sequential combining and averaging the characteristicsof the closest data of a limited sample, when it is not possible to put forward a hypothesisabout the initial number of clusters. Next, connect iterative clustering algorithms that providehigh stability with respect to noise features and the presence of outliers. Hybridization increasesthe efficiency of clustering algorithms. The research results showed that in order to increase thecomputational efficiency and overcome the sensitivity when initializing the parameters of clusteringalgorithms, it is necessary to use metaheuristic approaches to optimize the parameters of thelearning model and search for a global optimal solution.

Performance analysis in text clustering using k-means and k-medoids algorithms for Malay crime documents

<span lang="EN-US">Few studies on text clustering for the Malay language have been conducted due to some limitations that need to be addressed. The purpose of this article is to compare the two clustering algorithms of k-means and k-medoids using Euclidean distance similarity to determine which method is the best for clustering documents. Both algorithms are applied to 1000 documents pertaining to housebreaking crimes involving a variety of different modus operandi. Comparability results indicate that the k-means algorithm performed the best at clustering the relevant documents, with a 78% accuracy rate. K-means clustering also achieves the best performance for cluster evaluation when comparing the average within-cluster distance to the k-medoids algorithm. However, k-medoids perform exceptionally well on the Davis Bouldin index (DBI). Furthermore, the accuracy of k-means is dependent on the number of initial clusters, where the appropriate cluster number can be determined using the elbow method.</span>

Machine Learning-Based Rockfalls Detection with 3D Point Clouds, Example in the Montserrat Massif (Spain)

Rock slope monitoring using 3D point cloud data allows the creation of rockfall inventories, provided that an efficient methodology is available to quantify the activity. However, monitoring with high temporal and spatial resolution entails the processing of a great volume of data, which can become a problem for the processing system. The standard methodology for monitoring includes the steps of data capture, point cloud alignment, the measure of differences, clustering differences, and identification of rockfalls. In this article, we propose a new methodology adapted from existing algorithms (multiscale model to model cloud comparison and density-based spatial clustering of applications with noise algorithm) and machine learning techniques to facilitate the identification of rockfalls from compared temporary 3D point clouds, possibly the step with most user interpretation. Point clouds are processed to generate 33 new features related to the rock cliff differences, predominant differences, or orientation for classification with 11 machine learning models, combined with 2 undersampling and 13 oversampling methods. The proposed methodology is divided into two software packages: point cloud monitoring and cluster classification. The prediction model applied in two study cases in the Montserrat conglomeratic massif (Barcelona, Spain) reveal that a reduction of 98% in the initial number of clusters is sufficient to identify the totality of rockfalls in the first case study. The second case study requires a 96% reduction to identify 90% of the rockfalls, suggesting that the homogeneity of the rockfall characteristics is a key factor for the correct prediction of the machine learning models.

Application Of The Preference Selection Index And Method K-Means Clustering Method In Sales Product ( Case Study Of Ocha Cell Store)

The ocha cell shop is one of the shops engaged in the sale of mobile phones of various brands of mobile phones, the store has not used a special application in processing the products sold, but still uses manual data processing such as recording in a book. So that there is a accumulation of data that is not known for its benefits. One method that can solve this problem is the psi method and the k-means clustering method, because the psi method and the k-means clustering method are methods for solving multiple decision-making criteria, this method is useful if there is a conflict in determining the relative between attributes, so later on the final result of the calculation is that ranking is carried out by analyzing data on sales of goods that are in demand and not selling and the algorithm is partitional by using the value process repeatedly to get the value of the cluster database, it takes the desired number of initial clusters to produce the final ceteroid point as output. With this method, it helps store management to find out which cellphones are often sold and each time a transaction can later help in overcoming existing problems.

DP-k-modes: A self-tuning k-modes clustering algorithm

• Self-tuning cutoff distance is proposed to extend DPC. • Optimal initial seeds and number of clusters determined simultaneously. • New dissimilarity measure is proposed calculating dissimilarities between objects. • Self-tuning k-modes clustering algorithm , referred to as DP-k-modes, is proposed. • DP-k-modes is superior to the compared ones and is significant different to them. The k -modes clustering algorithm was proposed by Huang for handling datasets with categorical attributes, however, the dissimilarity measure used limits its applicability. Ng et al. improved on Huang’s k -modes algorithm by proposing a new dissimilarity measure between objects. Moreover, both k -modes algorithms require the initial seeds to be randomly chosen and the number of clusters be specified manually. To overcome the limitations of Huang’s and Ng’s k -modes clustering algorithms, we first extend the clustering algorithm published in Science in 2014 (“clustering by fast search and find of density peaks”). The optimal initial seeds and the number of clusters of a dataset are determined simultaneously by taking the standard deviation as the self-tuning cutoff distance and the simple match dissimilarity as the distance measurement in the definition of the density of a point. A new dissimilarity measure is proposed to calculate the dissimilarities between objects to improve on that of Ng’s k -modes algorithm. The performance of our resulting self-tuning k -modes clustering algorithm was tested on nine datasets (three being relatively large) from the UCI (University of California in Irvine) machine learning repository. The clustering results were compared to those produced by Huang’s and Ng’s algorithms. Statistical tests of three k -modes algorithms were undertaken to determine whether or not there is significant difference between our self-tuning k -modes algorithm and Huang’s and Ng’s k -modes algorithms. All these experimental results demonstrate that our proposed k -modes clustering algorithm is superior to Hang’s and Ng’s k -modes algorithms in terms of clustering accuracy ( ACC ) and the well-known Adjusted Rand Index ( ARI ) metric. Our self-tuning k -modes algorithm is significantly different from both Huang’s and Ng’s k -modes algorithms, and there is no statistically significant difference between Ng’s and Huang’s k -modes algorithms.

MetaConClust - Unsupervised Binning of Metagenomics Data using Consensus Clustering.

Background: Binning of metagenomic reads is an active area of research, and many unsupervised machine learning-based techniques have been used for taxonomic independent binning of metagenomic reads. Objective: It is important to find the optimum number of the cluster as well as develop an efficient pipeline for deciphering the complexity of the microbial genome. Methods: Applying unsupervised clustering techniques for binning requires finding the optimal number of clusters beforehand and is observed to be a difficult task. This paper describes a novel method, MetaConClust, using coverage information for grouping of contigs and automatically finding the optimal number of clusters for binning of metagenomics data using a consensus-based clustering approach. The coverage of contigs in a metagenomics sample has been observed to be directly proportional to the abundance of species in the sample and is used for grouping of data in the first phase by MetaConClust. The Partitioning Around Medoid (PAM) method is used for clustering in the second phase for generating bins with the initial number of clusters determined automatically through a consensus-based method. Results: Finally, the quality of the obtained bins is tested using silhouette index, rand Index, recall, precision, and accuracy. Performance of MetaConClust is compared with recent methods and tools using benchmarked low complexity simulated and real metagenomic datasets and is found better for unsupervised and comparable for hybrid methods. Conclusion: This is suggestive of the proposition that the consensus-based clustering approach is a promising method for automatically finding the number of bins for metagenomics data.

Inoculation of pear flowers with Metschnikowia reukaufii and Acinetobacter nectaris enhances attraction of honeybees and hoverflies, but does not increase fruit and seed set.

Currently, one of the most important challenges is to provide sufficient and affordable food and energy for a fast-growing world population, alongside preserving natural habitats and maintaining biodiversity. About 35% of the global food production depends on animals for pollination. In recent years, an alarming worldwide decline in pollinators has been reported, putting our food production under additional pressure. Therefore, there is an urgent need to find sustainable ways to ensure this crucial ecosystem service. Recent studies have shown that floral nectar is generally colonized by microorganisms, specifically yeasts and bacteria, which may alter nectar chemistry and enhance attraction of pollinators. In this study, we investigated changes in pollinator foraging behavior and pollination success in European pear (Pyrus communis L.) cultivars ‘Regal Red’ and ‘Sweet Sensation’ (red sports of ‘Doyenné de Comice’) after flower inoculation with the typical nectar-inhabiting microorganisms Metschnikowia reukaufii and Acinetobacter nectaris, and a combination of both. Pollination success was monitored by measuring the number of flower visits, fruit set and seed set in two consecutive years, 2019 and 2020. Results revealed that application of a mixture of M. reukaufii and A. nectaris resulted in significantly higher visitation rates of honeybees and hoverflies. By contrast, no effects on flower visits were found when yeasts and bacteria were applied separately. Fruit set and seed set were not significantly affected by any of the inoculation treatments. The only factors affecting fruit set were initial number of flower clusters on the trees and the year. The absence of treatment effects can most likely be attributed to the fact that pollination was not a limiting factor for fruit set in our experiments. Altogether, our results show that inoculation of flowers with nectar microbes can modify pollinator foraging patterns, but did not lead to increased pollination success under the conditions tested.

Incremental fuzzy clustering of time series

Clustering is one of the most popular data mining methods for analyzing the time series, not only due to its exploratory power, but also because it is often a preprocessing step or subroutine for other techniques. In this paper, an incremental fuzzy clustering algorithm of time series (IFCTS) is proposed, in which the clustering is divided into two stages: offline and online clustering. During the offline clustering process, the fuzzy clustering validity evaluation index is introduced to FCM to automatically obtain the optimal number of initial clusters for off-line data. Then, in the online clustering process, IFCTS algorithm can dynamically update the existing clusters in the incremental data snapshot, distinguish outliers and control the creation of new clusters adaptively by combining with the previous clusters. The experimental results show that the proposed algorithm has good clustering accuracy and efficiency for both equal-length and unequal-length time series.

Prediction of Environmental Pollution Using Hybrid PSO-K-Means Approach

Air pollution is increasing day by day, decreasing the world economy, degrading the quality of life, and resulting in a major productivity loss. At present, this is one of the most critical problems. It has a significant impact on human health and ecosystem. Reliable air quality prediction can reduce the impact it has on the nearby population and ecosystem; hence, improving air quality prediction is the prime objective for the society. The air quality data collected from sensors usually contains deviant values called outliers which have a significant detrimental effect on the quality of prediction and need to be detected and eliminated prior to decision making. The effectiveness of the outlier detection method and the clustering methods in turn depends on the effective and efficient choice of parameters like initial centroids and number of clusters, etc. The authors have explored the hybrid approach combining k-means clustering optimized with particle swarm optimization (PSO) to optimize the cluster formation, thereby improving the efficiency of the prediction of the environmental pollution.

Unsupervised color image segmentation: A case of RGB histogram based K-means clustering initialization.

Color-based image segmentation classifies pixels of digital images in numerous groups for further analysis in computer vision, pattern recognition, image understanding, and image processing applications. Various algorithms have been developed for image segmentation, but clustering algorithms play an important role in the segmentation of digital images. This paper presents a novel and adaptive initialization approach to determine the number of clusters and find the initial central points of clusters for the standard K-means algorithm to solve the segmentation problem of color images. The presented scheme uses a scanning procedure of the paired Red, Green, and Blue (RGB) color-channel histograms for determining the most salient modes in every histogram. Next, the histogram thresholding is applied and a search in every histogram mode is performed to accomplish RGB pairs. These RGB pairs are used as the initial cluster centers and cluster numbers that clustered each pixel into the appropriate region for generating the homogeneous regions. The proposed technique determines the best initialization parameters for the conventional K-means clustering technique. In this paper, the proposed approach was compared with various unsupervised image segmentation techniques on various image segmentation benchmarks. Furthermore, we made use of a ranking approach inspired by the Evaluation Based on Distance from Average Solution (EDAS) method to account for segmentation integrity. The experimental results show that the proposed technique outperforms the other existing clustering techniques by optimizing the segmentation quality and possibly reducing the classification error.

Effective noise reduction algorithm for material decomposition in dual-energy X-ray inspection

Dual-energy X-ray inspection can provide material-decomposed transmission information of the imaged object enhancing observer’s inspection performance. The material decomposition algorithms have an inherent noise amplification problem due to the multiplication and division operations during dual-energy image processing, which often degrades the material decomposition performance. In this work, we propose an effective and reliable clustering algorithm to resolve the noise amplification problem. The proposed method utilizes a k-means cluster algorithm as a base method with Gaussian noise of the data considered. In addition, we added a probabilistic regularization that can effectively relieve the decision challenge of the initial number of clusters. We have applied the proposed denoising method to the prototype X-ray dual-energy inspection system with an empirical calibration method that we have previously proposed. The proposed algorithm showed substantial reduction of noise in the test container projections and accordingly improved the material decomposition accuracy. Moreover, we performed quantitative analysis of the empirical calibration method with and without the cluster-based noise reduction algorithm and also on the conventional calibration method for material decomposition. It showed that the empirical calibration method with the noise reduction algorithm is robust against the thickness variation of the target object. The results suggest that the proposed noise reduction algorithm with the empirical calibration method is a promising solution to dual-energy X-ray material decomposition for inspection.

A Novel Clustering Algorithm Introducing New Denoising Technique

Breast cancer is one of common cancers in the developing countries. Detection at an early stage is very crucial for better chance of treatment. The techniques used to detect breast cancer are complex and time consuming. Computerized extraction of tumor areas from mammogram images is challenging due to shape and density of breast tumors which can sometimes surrounded by mucous (mucin). One of the challenges is to detect boundaries which can be blurred under noise factor. In this paper, we are introducing a clustering technique combined with specific structural features operations. A new noise elimination algorithm eases the noise problem and enhance the segmentation process using discrete cosine transform. Followed is the segmentation phase where classifying breast tumor from normal tumor are performed using a combined DCT and fuzzy c means algorithm. The contributions of the research are utilizing new filtering technique for noise removal. We also use Fuzzy C mean clustering algorithm using DCT information to determine the initial number of clusters. The tumor extracted segments are then transferred to the frequency domain using DCT and is used to for classifications. A test algorithm is implemented to classify new mammograms. Experimental results for all the proposed algorithms are extensively performed. The noise removal algorithm are proven robust. The experimental results of the search algorithm depicted different match and mismatch cases. 93% of the cases were a match case and predicted correctly. 5% were light cases and could not be detected from the images.

OCE-NGC: A neutrosophic graph cut algorithm using optimized clustering estimation algorithm for dermoscopic skin lesion segmentation

Automated skin lesion segmentation is one of the most crucial stages in dermoscopic images based diagnosis. To guarantee efficient unsupervised clustering-based segmentation, a histogram-based clustering estimation (HBCE) algorithm can be used to obtain the initial number of clusters with their corresponding centroids. Accordingly, the present work introduced a novel skin lesion segmentation algorithm, called optimized clustering estimation for neutrosophic graph cut algorithm (OCE-NGC). Firstly, the genetic algorithm (GA) is used to optimize the HBCE procedure by finding its optimal threshold values which are functions of a factor, called β to be optimized. This optimization process guarantees the optimal determination of the initial number of clusters and their corresponding centroids for further use in the proposed clustering process. Thus, the skin lesion dermoscopic images are then mapped into the neutrosophic set (NS) domain which is computed by the neutrosophic c-means (NCM). The NCM groups the pixels in the dermoscopic images using the pre-determined optimal number of clusters obtained by the optimized HBCE. Finally, a cost function of the graph cut (GC) algorithm is defined in the NS domain for the segmentation process. The experimental results established the superiority of the proposed OCE-NGC approach in comparison with the traditional HBCE with NCM only, the traditional HBCE with the NGC, and the typical GC. In a public dataset, the proposed approach achieved 97.12% and 86.28% average accuracy and average Jaccard (JAC) values, respectively.

Missing Data Imputation in High Dimensional Data Set using Local Similarity

Data quality is an important aspect for any data mining and statistical tasks. Presence of missing values in the dataset affects the data quality. Missing values refers to the event did not happen or the value does not exist. Data mining algorithms are not robust towards incomplete data. Imputation of missing values is necessary to improve the data quality for performing data mining and statistical analysis. The existing methods such as Expectation Maximization Imputation (EMI), A Framework for Imputing Missing values Using co appearance, correlation and Similarity analysis (FIMUS) use the whole dataset to impute missing values. In such cases, due to the influence of irrelevant record the accuracy of imputation may be affected. This can be controlled by only considering locally similar records to impute missing values. Local similarity imputation can be done through clustering algorithms such as k-means algorithm. K-means clustering efficiency depends on the number of clusters is to be defined by users. To increase the clustering efficiency, first distinctive value is imputed in place of missing ones and this imputed dataset is given to stacked autoencoder for dimensionality reduction which also improves the efficiency of clustering. Initial number of clusters to k-means algorithm is determined using fast clustering. Due to initial imputation, some irrelevant records may be partitioned to a cluster. When these records are used for imputing missing values, accuracy of imputation decreases. In the proposed algorithm, local similarity imputation algorithm uses only top knearest neighbours within the cluster to impute missing values. The performance of the proposed algorithm is evaluated based on Root-Mean-Squared-Error (RMSE) and Index of Agreement (d2). University of California Irvine datasets has been used for analyzing the performance of the proposed algorithm.