Two-stage Clustering Algorithm Research Articles

Optimizing Business-to-Business Customer Satisfaction Analysis through Advanced Two-Stage Clustering: Insights from Industrial Parks

Traditional research on customer satisfaction (CS) estimation has focused on the business-to-customer (BTC) business mode. Customers in the BTC mode have been assumed to be familiar with the full range of services or products and to be able to make estimations of their CS. However, in the business-to-business (BTB) mode, diverse services have often been required and provided. It may be difficult to find members who have experience with all kinds of services or to generate common CS estimation results supported by different members. In this study, the difference between BTC and BTB was verified using structural equation modeling (SEM), and a model of CS estimation was developed with respect to BTB. The empirical results show that perceived service quality has no direct impact on enterprise satisfaction, indicating that traditional models are limited. A two-stage clustering algorithm was adopted to optimize the traditional CS evaluation model based on SEM, i.e., (1) K-nearest neighbor (KNN) classification and (2) density-based spatial clustering of applications with noise (DBSCAN). In order to verify the feasibility of the proposed model, CS with respect to six industrial parks was estimated empirically. The results show that the proposed model can improve the results of CS estimation compared with the results obtained using traditional methods. During the clustering process, each park generated and eliminated a certain number of noise points to optimize the satisfaction evaluation results. Specifically, park A generated and eliminated seven noise points, while park C generated and eliminated five noise points. The results of the satisfaction evaluation of each park obtained using the proposed model are more realistic, i.e., park A > park B > park C > park E > park D > park F. The proposed model extends the existing research on CS estimation in theory and can support applications in the BTB business mode.

A new driving style recognition method for personalized adaptive cruise control to enhance vehicle personalization

There are limitations of personalization in Advanced Driver Assistance Systems (ADAS) that have a serious impact on driver acceptance and satisfaction. This study investigates driving style recognition method to achieve personalization of longitudinal driving behavior. Currently, driving style recognition algorithms for Personalized Adaptive Cruise Control (PACC) rely on integrated recognition. However, disturbances in the driving cycle may lead to changes in a driver’s integrated driving style. Therefore, the integrated driving style cannot accurately and comprehensively reflect the driver’s driving style. To solve this problem, a new driving style recognition method for PACC is proposed, which considers integrated driving style and driving cycle. Firstly, the method calculates the constructed feature parameters of driving cycle and style, and then reduces the dimensionality of the feature parameter matrix by principal component analysis (PCA). Secondly, a two-stage clustering algorithm with self-organizing mapping networks and K-means clustering (SOM-K-means) is used to obtain the type labels. Then, a transient recognition model based on random forest (RF) is established and the hyperparameters of this model are optimized by sparrow search algorithm (SSA). Based on this, a comprehensive driving style recognition model is established using analytic hierarchy process (AHP). Finally, the validity of the proposed method is verified by a natural dataset. The method incorporates the driving cycle into driving style recognition and provides guidance for improving the personalization of adaptive cruise control system.

Data-driven robust optimization for a multi-trip truck-drone routing problem

Using drones along with conventional vehicles, such as trucks, can potentially improve the cost-effectiveness and speed of delivery operations. This research considers a multi-trip truck-drone routing problem under uncertainty. Each truck is allowed to take multiple trips from the depot,while each drone is allowed to perform a single trip from the launch locations. Based on customer preferences, deliveries are grouped into deliveries by trucks to distribution centers and deliveries by drones from distribution centers to customers. A mixed-integer linear programming model is formulated to minimize the sum of the waiting times of all customers. To deal with uncertainty, a new two-stage clustering algorithm that uses multiple-kernel learning-based and single-kernel learning-based methods is introduced to construct uncertainty sets. In the case of column-wised uncertainty, a two-stage clustering with a dimensional separation algorithm is developedto avoid over-conservatism.A two-step solution method is provided for solving the proposed robust model to reduce the CPU time. The performance of the proposed algorithms is evaluated on several test problems. The results indicate that the developed algorithms can effectively avoid over-conservatism, reduce variables and constraints of the data-driven robust counterpart model, and reduce the CPU time required to solve the problem.

Comprehensive Identification of Driving Style Based on Vehicle's Driving Cycle Recognition

The driving style of human automobile drivers plays a vital role in vehicle energy management and driving safety. Human drivers often exhibit different driving styles during different driving cycles. Moreover, short-term driving behaviors cannot accurately reflect their comprehensive driving styles. Thus, a comprehensive driving style identification model, based on driving cycle recognition, is proposed in this study. The model was developed to comprehensively assess and more accurately identify long-term driving styles. First, the characteristic parameters of the driving cycle and style are selected and calculated, and the dimensions of the data collected from real vehicles are minimized through principal component analysis (PCA). Second, a two-stage clustering algorithm based on the particle swarm optimization K-means optimized by the genetic algorithm (GA-PSO-K-means) is used to obtain the type labels of driving cycles and styles to ensure the stability of the clustering algorithm. Third, an artificial neural network (ANN) is utilized to build the instantaneous recognition model of the driving cycle and style. Based on this, a comprehensive driving style identification model is built using the analytic hierarchy process(AHP). Finally, the bench and real vehicle tests are used to verify that the proposed model is accurate and effective.

Quantifying the comprehensive benefit of the bike-sharing system under cycling behavior differences

Bike-sharing (BS) as one of the sustainable travel modes, has become increasingly popular in cities around the world. However, the comprehensive benefit (CB) derived from multiple impacts in terms of environmental, time, convenience, and payment perspectives have not been well investigated. The study proposes an Environment-Time-Convenience-Payment (ETCP) model to quantitatively evaluate the CB of different clusters of BS users. More than 340,000 BS users in Beijing are clustered into 6 groups through a two-stage clustering algorithm, which are low-frequency travel users, subway-dependent users, less subway-dependent users, long-distance travel users, work purpose users, and BS preferred users. On this basis, 8 alternative scenarios of BS are suggested to address the mode substitution uncertainty. The study monetizes the CO2 emission and the travel-related time, so that various benefits can be quantified and estimated uniformly. The results show that long-distance travel users can bring positive CBs under all scenarios, with an average CB of 1.01 to 6.47 RMB per day for each user. Although BS usage could increase CO2 emissions under most scenarios, it would correspond to travel time or/and hassle time savings at the same time. This study also investigates the beneficial differences between dock-based and dock-less BSs existing in different cities. The dock-based system in Chicago could provide the best environmental benefit, while the dock-less one in Beijing is relatively lower due to its flexibility with more short-distance trips.

Improved Vessel Trajectory Prediction Model Based on Stacked-BiGRUs

An intelligent maritime navigation system is expected to play an important role in the realm of Internet of Vessels (IoV). As a key technology in navigation systems, vessel trajectory prediction technology is critical to the IoV. Automatic identification system (AIS), an automated tracking system, is used extensively for vessel trajectory prediction. However, certain characteristics in the AIS data, such as the large number of anchored trajectories in the area, anomalous sharp turns of some trajectories, and the behavioral differences of vessels in different segments, limit the prediction accuracy. In this study, we propose a novel vessel trajectory prediction model for accurate prediction with the following characteristics: (1) an anchor trajectory elimination algorithm to eliminate anchor trajectories; (2) a statistical trajectory restoration algorithm to repair sharp turning; (3) a two-stage clustering algorithm (D-KMEANS) to distinguish vessel behavior; and (4) a deep bidirectional gate recurrent unit (Stacked-BiGRUs) model to predict vessel trajectory and compare the accuracy of the model before and after improvement. The results show that the mean square error and the mean absolute error of the improved model are reduced by 27% and 46%, respectively. This research shows good potential for maritime navigation early warning and safety.

Heterogeneous cryo-EM projection image classification using a two-stage spectral clustering based on novel distance measures.

Single-particle cryo-electron microscopy (cryo-EM) has become one of the mainstream technologies in the field of structural biology to determine the three-dimensional (3D) structures of biological macromolecules. Heterogeneous cryo-EM projection image classification is an effective way to discover conformational heterogeneity of biological macromolecules in different functional states. However, due to the low signal-to-noise ratio of the projection images, the classification of heterogeneous cryo-EM projection images is a very challenging task. In this paper, two novel distance measures between projection images integrating the reliability of common lines, pixel intensity and class averages are designed, and then a two-stage spectral clustering algorithm based on the two distance measures is proposed for heterogeneous cryo-EM projection image classification. In the first stage, the novel distance measure integrating common lines and pixel intensities of projection images is used to obtain preliminary classification results through spectral clustering. In the second stage, another novel distance measure integrating the first novel distance measure and class averages generated from each group of projection images is used to obtain the final classification results through spectral clustering. The proposed two-stage spectral clustering algorithm is applied on a simulated and a real cryo-EM dataset for heterogeneous reconstruction. Results show that the two novel distance measures can be used to improve the classification performance of spectral clustering, and using the proposed two-stage spectral clustering algorithm can achieve higher classification and reconstruction accuracy than using RELION and XMIPP.

Two-stage clustering algorithm based on evolution and propagation patterns

Two-stage clustering algorithm based on evolution and propagation patterns

Correlation based neuro-fuzzy Wiener type wind power forecasting model by using special separate signals

The wind power system is characterized by complex structures, numerous effect elements, and complicated physical relationships, which bring great difficulties to establish the wind power forecasting model and result in the unsatisfactory prediction accuracy of wind power. In this paper, a correlation based neuro-fuzzy Wiener type wind power forecasting model is proposed by using special separate signals to decouple dynamic linear and static nonlinear characteristics of wind power systems for high-accuracy wind power forecasting. This method introduces an autoregressive exogenous model and fuzzy neural network to explore the underlying distribution characteristics of wind power systems by correlation analysis of the linear and nonlinear information contained in time-series data of the wind power system. In addition, in order to eliminate outliers in the original data set, an abnormal data cleaning approach combining quartile and two-stage clustering algorithm is employed to reduce the computation time and improve the model accuracy. Moreover, the convergence of the Wiener model is proved by the Lyapunov function. Simulation results verify the high accuracy and generalization ability of the proposed model both in the gale and breeze seasons.

A Modular Framework for Centrality and Clustering in Complex Networks

The structure of many complex networks includes edge directionality and weights on top of their topology. Network analysis that can seamlessly consider combination of these properties are desirable. In this paper, we study two important such network analysis techniques, namely, centrality and clustering. An information-flow based model is adopted for clustering, which itself builds upon an information theoretic measure for computing centrality. Our principal contributions include (1) a generalized model of Markov entropic centrality with the flexibility to tune the importance of node degrees, edge weights and directions, with a closed-form asymptotic analysis, which (2) leads to a novel two-stage graph clustering algorithm. The centrality analysis helps reason about the suitability of our approach to cluster a given graph, and determine ‘query’ nodes, around which to explore local community structures, leading to an agglomerative clustering mechanism. Our clustering algorithm naturally inherits the flexibility to accommodate edge directionality, as well as different interpretations and interplay between edge weights and node degrees. Extensive benchmarking experiments are provided, using both real-world networks with ground truth and synthetic networks.

On a two-stage progressive clustering algorithm with graph-augmented density peak clustering

Due to the rapidly growing volume and velocity of big data, real-time streaming data analysis has become increasingly important in many applications. To discover knowledge from such data, a wide range of machine learning techniques have been proposed and used in practice. Among them, clustering, which aims at grouping objects into different classes on the basis of their similarity, is the most common form of unsupervised learning. However, most existing clustering algorithms are designed for static data, and hence are not best suited for streaming data. In this paper, we propose PC-DPC, a two-stage progressive clustering algorithm with graph-augmented density peak clustering. PC-DPC first identifies clusters of streaming data using an improved density peak clustering algorithm, and then merges newly arriving data into the existing data pool by measuring inter-cluster structural similarity, which considers the distance between a center and representative points. We illustrate the superiority of PC-DPC over several state-of-the-art clustering algorithms in terms of clustering accuracy and running time on publicly available benchmark datasets.

Modeling storage systems in electricity markets with high shares of renewable generation: A daily clustering approach

In the energy transition towards a carbon-free society, the continuous changes that energy systems are experiencing, the increasing penetration of renewable generation, and the incorporation of short-term storage technologies such as batteries have increased the effort to model and predict its development and operation. In this context, power system models play a relevant role. However, to be suitable for the decision-making process, these tools should not require a massive computational effort. To overcome this challenge, this paper proposes a new methodology that reduces the temporal dimension of the problem while maintaining accurate results. This methodology is specially designed for medium-term operation of real-size power systems with a significant presence of renewable generation and storage systems. By means of a two-stage clustering algorithm, the proposed approach transforms the temporal structure of the model’s input parameters into different levels of time aggregation. This arrangement makes the problem manageable and computationally tractable. In addition, together with the newly incorporated model formulation, this methodology allows capturing at the same time the short- and medium-term variability present in power systems.

A Clustering Analysis Method With High Reliability Based on Wilcoxon-Mann-Whitney Testing

As a core step in clustering analysis, distance measurement results can influence clustering accuracy. Existing measurement methods are mostly based on cluster feature information. However, these cluster features may be insufficient and result in losing data information for clusters containing a number of objects. To improve measurement accuracy, we make full use of the distribution characteristics of objects in clusters, i.e., we use descriptive statistics and the Wilcoxon-Mann-Whitney rank sum test in nonparametric statistics to measure distances during clustering. Furthermore, we propose a two-stage clustering algorithm to improve clustering analysis performance. In terms of avoiding preliminarily assuming the number of clusters, with the proposed distance measurement method, the clustering algorithm can discover clusters with arbitrary shapes and improve clustering accuracy. Experiments with multiple datasets compared with other clustering algorithms illustrate the accuracy and efficiency of the proposed clustering algorithm.

An Effective Two-Stage Clustering Method for Mixing Matrix Estimation in Instantaneous Underdetermined Blind Source Separation and Its Application in Fault Diagnosis

The underdetermined blind source separation (UBSS) has been considered to be a novel signal processing technique, which can separate the fault source signals from their mixtures. The mixing matrix estimation is a major step in the UBSS, this paper focuses on boosting the accuracy level of the estimated mixing matrix in the underdetermined case. Since the traditional clustering algorithms may not capture the signal characteristics well and secure a satisfactory estimation of the mixing matrix, an effective two-stage clustering algorithm is proposed to estimate the mixing matrix through a combination of hierarchical clustering and K-means. More specifically, first, the sum of frequency points energy in the time-frequency (TF) domain is calculated to estimate the number of source signals before clustering, and the initial clustering centers are obtained with a hierarchical clustering algorithm. Second, after eliminating outliers deviating from the initial clustering centers with the cosine distance, the new clustering centers are obtained by recalculating the mean value of each sub-cluster. Finally, the new clustering centers are set as the initial clustering centers of the K-means algorithm to estimate the mixing matrix. Extensive simulations and experiments show that the proposed method can effectively separate the source signals and ensure an estimate of the mixing matrix that is substantially more accurate than the K-means algorithm alone.

A two-stage density clustering algorithm

Clustering by fast search and find of density peaks (CFDP) is a popular density-based algorithm. However, it is criticized because it is inefficient and applicable only to some types of data, and requires the manual setting of the key parameter. In this paper, we propose the two-stage density clustering algorithm, which takes advantage of granular computing to address the aforementioned issues. The new algorithm is highly efficient, adaptive to various types of data, and requires minimal parameter setting. The first stage uses the two-round-means algorithm to obtain $$\sqrt{n}$$ small blocks, where n is the number of instances. This stage decreases the data size directly from n to $$\sqrt{n}$$ . The second stage constructs the master tree and obtains the final blocks. This stage borrows the structure of CFDP, while the cutoff distance parameter is not required. The time complexity of the algorithm is $$O(mn^\frac{3}{2})$$ , which is lower than $$O (mn^2)$$ for CFDP. We report the results of some experiments performed on 21 datasets from various domains to compare a new clustering algorithm with some state-of-the-art clustering algorithms. The results demonstrated that the new algorithm is adaptive to different types of datasets. It is two or more orders of magnitude faster than CFDP.

Fast Clustering by Affinity Propagation Based on Density Peaks

Clustering is an important technique in data mining and knowledge discovery. Affinity propagation clustering (AP) and density peaks and distance-based clustering (DDC) are two significant clustering algorithms proposed in 2007 and 2014 respectively. The two clustering algorithms have simple and clear design ideas, and are effective in finding meaningful clustering solutions. They have been widely used in various applications successfully. However, a key disadvantage of AP is its high time complexity, which has become a bottleneck when applying AP for large-scale problems. The core idea of DDC is to construct the decision graph based on the local density and the distance of each data point, and then select the cluster centers, but the selection of the cluster centers is relatively subjective, and sometimes it is difficult to determine a suitable number of cluster centers. Here, we propose a two-stage clustering algorithm, called DDAP, to overcome these shortcomings. First, we select a small number of potential exemplars based on the two quantities of each data point in DDC to greatly compress the scale of the similarity matrix. Then we implement message-passing on the incomplete similarity matrix. In experiments, two synthetic datasets, nine publicly available datasets, and a real-world electronic medical records (EMRs) dataset are used to evaluate the proposed method. The results demonstrate that DDAP can achieve comparable clustering performance with the original AP algorithm, while the computational efficiency improves observably.

Clothescounter: A framework for star-oriented clothes mining from videos

This paper presents a novel framework, ClothesCounter, which aims to automatically identify clothes worn by certain stars in videos. At first, several deep convolutional neural networks (CNN) models were utilized to preprocess video data in order to detect clothing images from original video frames, including human body detection, human posture selection, human pose estimation, face verification, and clothing detection. We then propose a method for extracting features of clothing images based on triplet loss that can map clothing images into a compact feature space. In the learned feature space, we present a two-stage clustering algorithm that does not require the number of clusters. Our framework was examined in a large-scale video dataset. Experimental results demonstrate the feasibility and effectiveness of our proposed method.

An efficient space division–based width optimization method for RBF network using fuzzy clustering algorithms

The Radial Basis Function (RBF) network is one of the most widely used surrogate models in multidisciplinary design optimization. However, one of the challenges of applying the RBF network to engineering problems is how to efficiently optimize its width parameters. In this work, a novel space division–based width optimization (SDWO) method is proposed to decompose the complex multivariable width optimization into several small-scale single-variable width optimizations. The SDWO method consists of two main steps. First, a two-stage fuzzy clustering algorithm is carried out to group the samples and divide the input space into several overlapping local regions, and the overlapping degree is controlled by a dimensionless expansion factor. Second, in each local region, one small-scale local RBF network (LRBFN) is constructed by solving a single-variable optimization problem when the LRBFN is needed for prediction. All these LRBFNs are independent of each other and can be constructed in parallel. The proposed method is efficient and particularly suitable for large sample sets. Test results of four sample sets verify that the proposed SDWO method has better performance than the conventional width optimization method in terms of both training efficiency and approximation accuracy. An inter-stage structure optimization is carried out, which demonstrates the efficiency of the proposed method in practical engineering applications.

Adaptive Self-paced Deep Clustering with Data Augmentation

Deep clustering gains superior performance than conventional clustering by jointly performing feature learning and cluster assignment. Although numerous deep clustering algorithms have emerged in various applications, most of them fail to learn robust cluster-oriented features which in turn hurts the final clustering performance. To solve this problem, we propose a two-stage deep clustering algorithm by incorporating data augmentation and self-paced learning. Specifically, in the first stage, we learn robust features by training an autoencoder with examples that are augmented by random shifting and rotating the given clean examples. Then, in the second stage, we encourage the learned features to be cluster-oriented by alternatively finetuning the encoder with the augmented examples and updating the cluster assignments of the clean examples. During finetuning the encoder, the target of each augmented example in the loss function is the center of the cluster to which the clean example is assigned. The targets may be computed incorrectly, and the examples with incorrect targets could mislead the encoder network. To stabilize the network training, we select most confident examples in each iteration by utilizing the adaptive self-paced learning. Extensive experiments validate that our algorithm outperforms the state of the arts on four image datasets.

TripImputor: Real-Time Imputing Taxi Trip Purpose Leveraging Multi-Sourced Urban Data

Travel behavior understanding is a long-standing and critically important topic in the area of smart cities. Big volumes of various GPS-based travel data can be easily collected, among which the taxi GPS trajectory data is a typical example. However, in GPS trajectory data, there is usually little information on travelers’ activities, thereby they can only support limited applications. Quite a few studies have been focused on enriching the semantic meaning for raw data, such as travel mode/purpose inferring. Unfortunately, trip purpose imputation receives relatively less attention and requires no real-time response. To narrow the gap, we propose a probabilistic two-phase framework named TripImputor , for making the real-time taxi trip purpose imputation and recommending services to passengers at their dropoff points. Specifically, in the first phase, we propose a two-stage clustering algorithm to identify candidate activity areas (CAAs) in the urban space. Then, we extract fine-granularity spatial and temporal patterns of human behaviors inside the CAAs from foursquare check-in data to approximate the priori probability for each activity, and compute the posterior probabilities (i.e., infer the trip purposes) using Bayes’ theorem. In the second phase, we take a sophisticated procedure that clusters historical dropoff points and matches the dropoff clusters and CAAs to immerse the real-time response. Finally, we evaluate the effectiveness and efficiency of the proposed two-phase framework using real-world data sets, which consist of road network, check-in data generated by over 38 000 users in one year, and the large-scale taxi trip data generated by over 19 000 taxis in a month in Manhattan, New York City, USA. Experimental results demonstrate that the system is able to infer the trip purpose accurately, and can provide recommendation results to passengers within 1.6 s in Manhattan on average, just using a single normal PC.