Cluster Centroids Research Articles

결측치를 포함한 데이터의 k-평균 군집분석 방법 비교

Cluster analysis is an unsupervised learning method to find heterogeneous clusters that capture similarities among items and separate different items into different clusters. Various cluster analysis techniques have been proposed, and the k-means clustering method, which minimizes the sum of Euclidean distances between cluster centroids and individual entities, is widely recognized as a standard cluster analysis method. When data include missing values, it is challenging to conduct cluster analysis, because it is impossible to calculate distances between centroids of clusters and incomplete items, resulting in excluding classification of these items. Techniques have been suggested to handle missing values in k-means clustering, including conducting cluster analysis after imputation of missing values or cluster analysis based on available information. In this study, we explore methods to perform k-means cluster analysis on data with missing values and evaluate performance of these methods using a simulation. The results of simulation studies indicate that conducting k-means cluster analysis after imputation yields the better performance than the one based on available information. Among the various imputation methods, k-nearest neighbors imputation performed the best.

Adaptive K-means clustering based under-sampling methods to solve the class imbalance problem

In the field of machine learning, the issue of class imbalance is a common problem. It refers to an imbalance in the quantity of data collected, where one class has a significantly larger number of data compared to another class, which can negatively affect the classification efficiency of algorithms. Under-sampling methods address class imbalance by reducing the quantity of data in the majority class, thereby achieving a balanced dataset and mitigating the class imbalance problem. Traditional under-sampling methods based on k-means clustering either set the unified value of k (number of clusters) or determine it directly based on the quantity of data in the minority or majority class. This paper proposes an adaptive k-means clustering under-sampling algorithm that calculates an appropriate k for each dataset. After clustering the majority class dataset into k clusters, our algorithm calculates the distances between the data within each cluster and the cluster centroids from two perspectives and selects data based on these distances. Subsequently, the subset of the majority class dataset are combined with the minority class dataset to generate a new balanced dataset, which is then used for classification algorithms. The performance of our algorithm is evaluated on 45 datasets. Experimental results demonstrate that our algorithm can dynamically determine appropriate k for different datasets and output a balanced dataset, thus enhancing the classification efficiency of machine learning algorithms. This work can provide new algorithmic ensemble strategies for addressing class imbalance problem.

Improved Test Input Prioritization Using Verification Monitors with False Prediction Cluster Centroids

Deep learning (DL) systems have been remarkably successful in various applications, but they could have critical misbehaviors. To identify the weakness of a trained model and overcome it with new data collection(s), one needs to figure out the corner cases of a trained model. Constructing new datasets to retrain a DL model requires extra budget and time. Test input prioritization (TIP) techniques have been proposed to identify corner cases more effectively. The state-of-the-art TIP approach adopts a monitoring method to TIP and prioritizes based on Gini impurity; one estimates the similarity between a DL prediction probability and uniform distribution. This letter proposes a new TIP method that uses a distance between false prediction cluster (FPC) centroids in a training set and a test instance in the last-layer feature space to prioritize error-inducing instances among an unlabeled test set. We refer to the proposed method as DeepFPC. Our numerical experiments show that the proposed DeepFPC method achieves significantly improved TIP performance in several image classification and active learning tasks.

Semantic part segmentation of spatial features via geometric deep learning for automated control cabinet assembly

AbstractIndustries with batch size one manufacturing philosophies for highly customized products have been largely limited in manufacturing automation. The control cabinet industry is particularly affected by this problem during the mounting and wiring of components due to high variety, variance, and complexity of components as well as handling tasks. Rapid advances in the field of machine learning are opening new possibilities for automating previously manual processes. This paper proposes a concept for identifying geometric features of electrical components that starts from STEP files and transforms them into modular metrics relevant to build a digital twin and (automatic)manufacturing. The architecture is tested on a self-aggregated and processed dataset of control cabinet components and achieves an average dice score of 65.27% and an intersection over union of 51.41% across all segmentation classes. In addition to semantic part segmentation of the components, the cluster, volume and surface centroids, the normal vectors and the size of each feature are computed. The paper evaluates the suitability of cutting-edge techniques such as diffusion as well as established deep learning architectures. The result is a hybrid end-to-end inference pipeline suitable for general spatial assembly processes.

Stacking-ac4C: an ensemble model using mixed features for identifying n4-acetylcytidine in mRNA.

N4-acetylcytidine (ac4C) is a modification of cytidine at the nitrogen-4 position, playing a significant role in the translation process of mRNA. However, the precise mechanism and details of how ac4C modifies translated mRNA remain unclear. Since identifying ac4C sites using conventional experimental methods is both labor-intensive and time-consuming, there is an urgent need for a method that can promptly recognize ac4C sites. In this paper, we propose a comprehensive ensemble learning model, the Stacking-based heterogeneous integrated ac4C model, engineered explicitly to identify ac4C sites. This innovative model integrates three distinct feature extraction methodologies: Kmer, electron-ion interaction pseudo-potential values (PseEIIP), and pseudo-K-tuple nucleotide composition (PseKNC). The model also incorporates the robust Cluster Centroids algorithm to enhance its performance in dealing with imbalanced data and alleviate underfitting issues. Our independent testing experiments indicate that our proposed model improves the Mcc by 15.61% and the ROC by 5.97% compared to existing models. To test our model's adaptability, we also utilized a balanced dataset assembled by the authors of iRNA-ac4C. Our model showed an increase in Sn of 4.1%, an increase in Acc of nearly 1%, and ROC improvement of 0.35% on this balanced dataset. The code for our model is freely accessible at https://github.com/louliliang/ST-ac4C.git, allowing users to quickly build their model without dealing with complicated mathematical equations.

Wing morphometry of worker bees of Kharkiv, Sumy and Poltava regions of Ukraine

The affiliation of local bee populations spread across the territory of Ukraine to the subspecies A. m. carnica, A. m. macedonica, A. m. mellifera, and A. m. caucasica has been studied previously in most cases using various phenotypic characteristics, and in some cases, through the study of mitochondrial DNA. According to previous morphometric studies, it was not possible to identify bee families that could be qualified as "purebred". The aim of this study was to investigate three apiaries from different regions of Ukraine where selection measures are systematically carried out to reproduce and preserve the local population of Ukrainian steppe bees, subspecies A. m. macedonica. The goal was to confirm this breeding positioning and find bee families suitable for further breeding work. The study focused on the wing phenotypes of worker bees. In this work, 10 traits were considered as wing phenotypes: seven indices Ci, Dbi, Disc.sh, Ri, Ci.2, Ci.3, Ci.2.1, and three angles A4, E9, J10. Statistical data processing methods and the STATISTICA software package were used as research tools. A total of 5200 wings of worker bees from 53 apiaries in the Kharkiv, Sumy, and Poltava regions were studied. The classification of wings in the Sumy and Poltava regions was carried out into four clusters, while in Kharkiv, it was into three clusters. Preliminary classification based on Euclidean distances using the mean values of traits for individual wing clusters and taking into account the Mahalanobis distances between cluster centroids formed three data arrays labeled as "UkrStep" and "UkrStep.2," which were attributed to the subspecies A. m. macedonica, and "Carnica," attributed to the subspecies A. m. carnica. These can be used as regional standards for studying bee families in this part of Ukraine with the aim of establishing their probable breed. It was found that the population of Ukrainian steppe bees (62.2%) predominates by phenotype on the studied apiaries. For four bee families from the Kuzemin apiary, 92%-100% of worker bees belong to Ukrainian steppe bees. АOn the apiary in Kharkiv, only one bee family had 86% of worker bees belonging to the population of Ukrainian steppe bees. It is recommended to use queens from four bee families for breeding purposes: one queen from the Kharkiv apiary and three queens from the Kuzemin apiary. Key words: Classical morphometry of worker bee wings, discriminant analysis.

A-MKMC: An effective adaptive-based multilevel K-means clustering with optimal centroid selection using hybrid heuristic approach for handling the incomplete data

In general, clustering is defined as partitioning similar and dissimilar objects into several groups. It has been widely used in applications like pattern recognition, image processing, and data analysis. When the dataset contains some missing data or value, it is termed incomplete data. In such implications, the incomplete dataset issue is untreatable while validating the data. Due to these flaws, the quality or standard level of the data gets an impact. Hence, the handling of missing values is done by influencing the clustering mechanisms for sorting out the missing data. Yet, the traditional clustering algorithms fail to combat the issues as it is not supposed to maintain large dimensional data. It is also caused by errors of human intervention or inaccurate outcomes. To alleviate the challenging issue of incomplete data, a novel clustering algorithm is proposed. Initially, incomplete or mixed data is garnered from the five different standard data sources. Once the data is to be collected, it is undergone the pre-processing phase, which is accomplished using data normalization. Subsequently, the final step is processed by the new clustering algorithm that is termed Adaptive centroid based Multilevel K-Means Clustering (A-MKMC), in which the cluster centroid is optimized by integrating the two conventional algorithms such as Border Collie Optimization (BCO) and Whale Optimization Algorithm (WOA) named as Hybrid Border Collie Whale Optimization (HBCWO). Therefore, the validation of the novel clustering model is estimated using various measures and compared against traditional mechanisms. From the overall result analysis, the accuracy and precision of the designed HBCWO-A-MKMC method attain 93 % and 95 %. Hence, the adaptive clustering process exploits the higher performance that aids in sorting out the missing data issuecompared to the other conventional methods.

Optimizing Workload Scheduling in Cloud Paradigm using Robust Neutrosophic C-Means Clustering Boosted with Fish School Search

In the present internet world, accessing cloud resources for a low cost, according to their needs, is available to all users. Sharing resources is becoming increasingly necessary as people complete their activities in the cloud. It becomes essential for distributed workloads to be optimized to perform efficient workload scheduling and progressing resource utilization in a cloud environment. Scheduling cloud resources considerably benefits from the invention of machine learning and metaheuristic models to address this scenario. Though many existing algorithms are developed in cloud-based task scheduling using unsupervised clustering methods, the problem of unknown task requirements or resource availability in adverse conditions is still challenging. In this study, an uncertainty-based unsupervised technique is constructed to group incoming tasks according to the required resources, and it is scheduled to the most suitable resources more prominently. This paper introduced a Robust Neutrosophic C-Means Clustering boosted with the fish school search algorithm (RNCM-FSSA) for clustering the incoming tasks and the resources based on their requirement and availability. With the degree of indeterminacy, neutrosophic C-means discriminating the deterministic and indeterministic schemes and scheduling them to the optimal resources more effectively. Using the fitness value computed by FFSA, the potential cluster centroids are utilized for clustering, thus avoiding the early convergence in the grouping process. The simulation results explore that the robustness of the proposed RCNM-SSA achieves better resource utilization, the degree of imbalance is minimal, and computation complexity is also considerably decreased compared with other unsupervised models.

Ship route simulation based on the cluster analysis

The problem of safe ship’s route planning at the variable hydrometeorological situation along the route is considered in the paper. The problem solved in the paper is based on the division of the water area into separate clusters, depending on its characteristics. The route from Busan Port to Kushiro Port is used as an example. The specifics of this route is that it runs through the open sea, and through the straits and archipelagos. The aim of the work is to automate the process of planning the route and adjusting it during the trip, depending on changes in external conditions. The graph theory for modeling a route is proposed in this paper. The construction of graphs is implemented using the cluster analysis method. In the analysis, the water area of the route is divided into separate subareas, depending on the distance to the coast and the depths difference. Open water areas are separated into larger clusters. Near the coast and at shallow depths, clustering is shorter. The cluster centroids are the vertices of the graphs of the future route. As result it is a simulation of graphs of different sizes. The union of graphs forms the hypergraph. The distance is used as the weight of the graph edges. To determine the most preferable route in criteria of speed, cost or safety, weights are added depending on weather and other factors. As a result of this approach, several routes are planning on the hypergraph. Depending on the weight of the priority criteria, a route is selected automatically. The proposed method can be used to create systems for automated planning of optimal routes, taking into account the selected criteria under changing environmental conditions in the process of voyage and replanning it if necessary.

Two Dimensional Vortex Shedding from a Rotating Cluster of Cylinders

The dynamics of two-dimensional vortex shedding from a rotating cluster of three cylinders was investigated using Computational Fluid Dynamics (CFD) and Dynamic Mode Decomposition (DMD). The cluster was formed from three circles with equal diameters in mutual contact and allowed to rotate about an axis passing through the cluster centroid. While immersed in an incompressible fluid with Reynolds number of 100, the cluster was allowed to rotate at non-dimensionalised rotation rates (Ω) between 0 and 1. The rotation rates were non-dimensionalised using the free-stream velocity and the cluster characteristic diameter, the latter being equal to the diameter of the circle circumscribing the cluster. CFD simulations were performed using StarCCM+. Dynamic Mode Decomposition based on the two-dimensional vorticity field was used to decompose the field into its fundamental mode-shapes. It was then possible to relate the mode-shapes to lift and drag. Transverse and longitudinal mode-shapes corresponded to lift and drag, respectively. Lift–drag polars showed a more complex pattern dependent on Ω in which the flow fields could be classified into three regimes: Ω less than 0.3, greater than 0.5, and between 0.3 and 0.5. In general, the polars formed open curves in contrast to those of static cylinders, which were closed. However, some cases, such as Ω = 0.01, 0.22, and 0.28, formed closed curves. Whether a lift-drag polar was closed or open was deduced to be determined by the ratio of Strouhal numbers calculated using lift and drag time series, with closed curves forming when the ratio is an integer.

SAR model for accurate detection of multi-label arrhythmias from electrocardiograms

ObjectiveArrhythmias are prevalent symptoms of cardiovascular disease, necessitating accurate and timely detection to mitigate associated risks. Detecting arrhythmias from ECGs quickly and accurately holds great significance in preventing heart disease and reducing mortality. This research endeavors to outperform previous studies by developing a scientific neural network model capable of training and predicting ECG signals for 11 categories of arrhythmias, accounting for up to 5 co-existing labels. MethodsIn this study, we initially address the issue of imbalanced datasets by employing Borderline SMOTE and Cluster Centroids techniques during preprocessing. Subsequently, we propose a novel SAR model that combines attention and resnet mechanisms. The dataset is subjected to a 10-fold validation process to train and evaluate the model. Finally, several metrics such as HammingLoss, RankingLoss, F1-score, AUC and Coverage are used to evaluate the model. ResultsBy evaluating the results of the tests, the average Hamming Loss is 1.12 %, the average Ranking Loss is 1.17 %, the average Micro F1-score is 98.46 %, the average Micro AUC is 98.76 %, and the average Coverage is 3.2762. The results show that the SAR model outperforms previous related studies on the task of classifying arrhythmia signals with multiple categories and labels. ConclusionThe SAR model demonstrated excellent performance in accurately classifying multi-category and multi-label arrhythmia signals, affirming its scientific validity. Compared with previous studies, the model achieves a certain improvement in performance, which can help cardiologists to achieve scientific and accurate diagnosis of arrhythmia diseases.

Modular stochastic configuration network based prediction model for NOx emissions in municipal solid waste incineration process

The accurate prediction of the nitrogen oxides (NOx) emissions is extremely important for pollutant control in municipal solid waste incineration (MSWI) process. Modular neural network (MNN) provides a support for predicting NOx emissions to overcome the limitations of a single model in nonlinear processes and different climatic conditions. However, the design of sub model for MNN is a challenge. We propose a method based on MNN and adaptive ensemble stochastic configuration network (m-AESCN) in this work. First, fuzzy C-means algorithm decomposes the task into several sub datasets with similar characteristics, and an evaluation function is used to guarantee the optimal decomposition result. Second, for each sub dataset, an adaptive ensemble stochastic configuration network as the sub model of modular neural network. The optimal output for sub model can be obtained by an adaptive weighting method based on variance. Third, in testing phase, a sub model activation method based on cluster centroid is proposed to select a suitable sub model. Besides, the m-AESCN method is validated by the real data of an MSWI plant and shows considerable performance in two different datasets. And the proposed method is compared with several modeling methods, such as stochastic configuration network, support vector regression, random forest and so on. The experimental results under two datasets prove that the maximum improvements in accuracy of m-AESCN are 39.74% and 31.67%, respectively.

The influences of eccentricity on the fastener load and bearing strength of eccentric connection in aircraft structure

Purpose This paper aims to study the influences of eccentricity on the fastener load and bearing strength of the eccentric connection in the aircraft structure. Design/methodology/approach The special experiment is designed for the researches. The fastener loads of the eccentric connection are gained by using the derived formulas and numerical analysis, and the fastener load rules is verified by the experiment. The bearing strength of the eccentric connection is investigated by the experiments under different eccentricities compared with that gained from the experiment. Findings The study results are summarized as follows. Magnitude of the fastener load in the eccentric connection is greatly affected by distance from the fastener to the centroid of the fastener cluster and that from the fastener to the concentrated load. With the increase of eccentricity of the homolateral concentrated load, the fastener load increases, and difference of the fastener loads becomes larger, forming the short plate effect of the bucket. It means that fastener with the maximum load (the shortest plate of the bucket) leads to decrease of the bearing strength of the eccentric connection (the capacity of the bucket). Originality/value The investigation on the influence of eccentricity on the bearing strength of eccentric connection is firstly presented. The vector expression of the fastener load in eccentric connection is firstly derived. And the influencing mechanism of the fastener load on the bearing strengths of the different eccentric connections is demonstrated. The study results can provide guidance for the structure design of the eccentric connection.

Cancer Disease Treatment Classification Using Extended Rough based Intuitionistic Fuzzy C-means Learning Methodology

Because of the sensitivity of magnetic resonance to noise for the diagnosis and study of cancer, classifying magnetic resonance disease treatments is a medically complex and vitally important process. To sum up, these are the most pressing problems with the current system for describing how to treat cancer. Mathematical methods like rough sets, fuzzy sets, and rough sets are utilized to assess and cope with the ambiguity and uncertainty present in medical cancer disease treatments. Rough sets and rough sets-based techniques have both been offered in past, but depending on the parameters, each has its own special set of issues. The Extended Rough based Intuitionistic Fuzzy C-means Learning Technique (ERIFCM) with estimation of weight bias parameter for Cancer Disease Treatment Classification presented in this work is a novel method for computing the disease treatment classification. Using the non-membership and membership values in intuitionistic Rough based fuzzy sets, a generalised kind of fuzzy, rough sets and their representative components are evaluated. The approach that is proposed in this research uses existing clustering's standard features to lower the intensity and noise associated with cancer treatments without the need for spatial weight context data. In addition, the cluster centroid is initialized using the max-dist evaluation method, which is based on the weight measure, before the proposed algorithm is executed. This helps to reduce the number of iterations required for clustering. In contrast to existing segmented approaches developed for Cancer Disease Treatment and similar Disease Treatments, experimental results of the proposed approach yielded effective Disease Treatment Classification results. The main component of the proposed method is a more thorough analysis of experimental results.

Analysis of the Influence Path of Confucianism in the Civic Education of Contemporary College Students in the Context of Big Data

Abstract Under the background of big data, college students’ Civic Education has realized the adjustment of teaching technology and educational paradigm, and the value of big data technology for college students’ Civic Education has become increasingly intuitive. In this paper, firstly, the data points designated for clustering are selected in the dataset as the initial clustering centroids. In response to the Euclidean distance treating variables equally, the coefficient of variation method is proposed for subjective weighting. The support vector machine for solving the two-class classification problem is obtained as a real-valued function to test the parallel efficiency of different clustering mining algorithms in the same experimental setting. Finally, for the training sample input data, it is concluded that the algorithm has a better clustering effect and improves the speed and accuracy of clustering. The feasibility of the proposed influence path is demonstrated by comparing it with the data application of the subjective empirical weighting method. The results show that 8% of students before the experiment thought that integrating Confucianism into Civic Education was valuable and meaningful, and after the experiment, the number of students who thought it was valuable and meaningful increased from 8% to 58%. It indicates that the integration of excellent traditional Confucianism into the ideological and political education of college students is very important to strengthen the ideological and political education of college students.

GWOKM: A novel hybrid optimization algorithm for geochemical anomaly detection based on Grey wolf optimizer and K-means clustering

Identifying the geochemical signatures of valuable mineral deposits using regional geochemical data from stream sediments is a challenging task due to the intricate characteristics of geological formations. Our team is currently investigating the potential of unsupervised clustering analysis (CA) and hybridization with the grey wolf optimizer (GWO) in developing multi-element geochemical models using stream sediment data. To cluster the geochemical data and uncover any unusual patterns, we opted to use the K-means (KM) algorithm due to its straightforward implementation, fast computation speed, and capacity to handle the large datasets. Despite its benefits, the KM method also has notable limitations, such as the random selection of cluster centroids. This can result in higher systematic uncertainty in unsupervised geochemical modeling and longer computation times. To mitigate this concern, we have introduced a new hybrid approach, grey wolf optimizer with K-means so-called the GWOKM algorithm to enhance the delineation of multi-elemental patterns in stream sediment geochemical data. This method incorporates the grey wolf optimization algorithm with KM to optimize the identification of both anomalies and backgrounds using factor analysis and sample catchment basin modeling techniques. This approach was utilized to detect anomalous multi-elemental geochemical patterns indicative of porphyry and skarn copper deposits in the Baft area, Kerman belt, Iran. Upon comparison of the geochemical models derived from KM and GWOKM clustering methods, the latter outperformed the former, as evidenced by its higher prediction rate. The outcomes affirm the efficacy of unsupervised KM clustering analysis (CA) as a means of breaking down geochemical anomaly-background populations. Moreover, the integration of clustering methods with optimization algorithms has proven to be successful for enhancing the credibility of mineralized areas, which could be advantageous in future exploration phases. Based on the results, the GWOKM approach generates more reliable and efficient geochemical anomaly targets.

Research on unmanned transfer vehicle path planning for raw grain warehousing

A large number of grain machinery and vehicle equipment are usually required in the raw grain storage phase, and these objects together form the path planning map environment for the unmanned grain transfer vehicle. After using LiDAR to build a map of the environment for path planning, these dense and cluttered obstacles tend to affect the path planning effect making the unmanned transfer vehicle create a crossing from the impenetrable dense obstacles. To address this problem, this paper firstly deals with obstacles by fusing the DBSCAN clustering algorithm and K-means clustering algorithm, clustering obstacles, and extracting the cluster centroid and boundary points of each obstacle class to avoid the above situation. Secondly, the specific A* algorithm is improved, the search field way of the A* algorithm is optimized, and the optimized 5×5 field search way is used instead of the traditional 3×3 field search way of A* to improve the node search efficiency of the algorithm. Finally, the repulsion function of the artificial potential field algorithm is added to the A* heuristic function as a safety function to increase the obstacle avoidance capability of the A* algorithm. After verification, the improvement can operate better in the dense and cluttered obstacle environment.

Graph Laplacian-based spectral multi-fidelity modeling

Low-fidelity data is typically inexpensive to generate but inaccurate, whereas high-fidelity data is accurate but expensive. To address this, multi-fidelity methods use a small set of high-fidelity data to enhance the accuracy of a large set of low-fidelity data. In the approach described in this paper, this is accomplished by constructing a graph Laplacian from the low-fidelity data and computing its low-lying spectrum. This is used to cluster the data and identify points closest to the cluster centroids, where high-fidelity data is acquired. Thereafter, a transformation that maps every low-fidelity data point to a multi-fidelity counterpart is determined by minimizing the discrepancy between the multi- and high-fidelity data while preserving the underlying structure of the low-fidelity data distribution. The method is tested with problems in solid and fluid mechanics. By utilizing only a small fraction of high-fidelity data, the accuracy of a large set of low-fidelity data is significantly improved.

Feature Selection Based on Unsupervised Clustering Mechanism on Multiple-Sequence MRIs for Predicting Neoadjuvant Chemoradiation Response in Locally Advanced Rectal Cancer

Accurate response prediction allows for personalized cancer management. We developed an unsupervised clustering mechanism to improve effectiveness and efficiency in feature selection operation for accurate patient stratification. Forty-three locally advanced rectal cancer (LARC) patients underwent neoadjuvant chemoradiation were included, pre-treatment T2 and ADC MRIs were acquired for each patient. An initial feature space consisting of 200 radiomic features extracted from manually delineated GTVs from two sequences of MR images. Additional 960 high-order radiomic features extracted from a 3D convolutional neural network (CNN). To remove redundant and irrelevant features, we developed an unsupervised clustering-based feature selection operation to determine the combination of features with potential best performance. The normal process of feature selection involves searching new feature combinations and training new classifiers for evaluating their performance via an iterative process based on selected feature set, the overall time cost is tremendous. To balance the computational cost and search efficiency, firstly, we proposed an unsupervised clustering analysis metric- Comprehensive Cluster Analysis Index (CCAI) through the K-means algorithm, where the average distances between the sample points and the cluster centroids and so on, to construct a multiple linear regression model. Secondly, we extracted sample points by varying the number of features and feature ratios between radiomic features and 3D-CNN features in the output of feature selection. Thirdly, we optimized the model using the sampling points to calculate the CCAI. Two typical feature combination search algorithms, the random forest recursive feature elimination (RF-RFE) and the differential evolution (DE), were used to perform feature selection with CCAI. The accuracy, area-under-curve (AUC) and specificity, based on combined 3D-CNN and radiomic features extracted from combined T2 and ADC images, were 0.852, 0.871, and 0.735, respectively. Our experiments illustrated higher predictive power (AUC = 0.846) based on high-order abstract features extracted from the CNN on ADC and T2 images, compared to the traditional radiomic model (AUC = 0.714). Additionally, the predictive models constructed based on radiomics and CNN features extracted from ADC images were more predictable in terms of treatment responses than the radiomic and CNN imaging features extracted from T2 images. The average computational time of DE and RF-RFE were 50.5s and 128.6s in one single computation, the average computational time were 24.2s and 91.3s with CCAI, respectively. We proposed an unsupervised clustering analysis mechanism to improve the effectiveness of feature selection while decreasing its time cost markedly, which highlight the correlation and complementarity between low- and high-level imaging features, achieving better predictive accuracy.

Enhanced Robust Fuzzy K-Means Clustering joint [formula omitted]-norm constraint

Clustering is an unsupervised classical data processing technique, in which Fuzzy K-Means is extensively researched in practical application owing to its efficiency. However, common outliers in real-world always lead to clustering degradation. In this paper, we design an Enhanced Robust Fuzzy K-Means clustering method (ERFKM) joint ℓ0-norm sparse nearest neighbor constraint as a countermeasure to above challenge. On one hand, we adaptively weigh different samples by binary vector to measure each contribution for robustness, among which the outliers can be effectively recognized to eliminate their impact on clustering. That is, the weighted term is able to steadily correct the cumulative drift deviation of cluster centroid via ignoring extremely large sample-centroid residuals from outliers. On the other hand, the fuzzy regularizer with ℓ0-norm constraint aims to further enhance the segmentation ability of indicated normal samples through automatic cluster neighbor assignment. Finally, an optimization algorithm for our model is devised to simultaneously update the binary weight of each sample and sparse membership matrix for achieving robust clustering. Comprehensive experiments demonstrate the advantages and effectiveness of the proposed method.