Stream Clustering Research Articles

Adaptive Clustering for Dynamic IoT Data Streams

The emergence of the Internet of Things (IoT) has led to the production of huge volumes of real-world streaming data. We need effective techniques to process IoT data streams and to gain insights and actionable information from real-world observations and measurements. Most existing approaches are application or domain dependent. We propose a method which determines how many different clusters can be found in a stream based on the data distribution. After selecting the number of clusters, we use an online clustering mechanism to cluster the incoming data from the streams. Our approach remains adaptive to drifts by adjusting itself as the data changes. We benchmark our approach against state-of-the-art stream clustering algorithms on data streams with data drift. We show how our method can be applied in a use case scenario involving near real-time traffic data. Our results allow to cluster, label, and interpret IoT data streams dynamically according to the data distribution. This enables to adaptively process large volumes of dynamic data online based on the current situation. We show how our method adapts itself to the changes. We demonstrate how the number of clusters in a real-world data stream can be determined by analyzing the data distributions.

TAILS FROM THE ORPHANAGE

ABSTRACT Examining a portion of the northern Sloan Digital Sky Survey (SDSS) footprint, we detect at least three and possibly seven halo debris streams. One of these (PS1-D) was recently detected in the Pan-STARRS1 survey, and the remaining two are also evident as extensions of the SDSS detections. All of these streams are metal-poor and are found at a distance of around 21 ± 5 kpc. The streams are between 65° and 70° in length, oriented almost north–south, and are nearly parallel and somewhat convergent with the neighboring Orphan stream. Surface densities ranging from 1.5 to 0.5 stars per square degree down to correspond to surface brightnesses between 35 and 37 mag per square arcsecond. The streams each appear to be more than 300 pc across, suggesting either dwarf/ultrafaint galaxy progenitors or long-term heating of very ancient globular cluster streams. The orbits of all but one of these streams appear to be nearly radial, and the orbit normals suggest that all of the streams are part of the Vast Polar Structure, a relatively narrow plane that contains most of the known satellite galaxies, globular clusters, and stellar streams.

Introduction to stream: An Extensible Framework for Data Stream Clustering Research with R

In recent years, data streams have become an increasingly important area of research for the computer science, database and statistics communities. Data streams are ordered and potentially unbounded sequences of data points created by a typically non-stationary data generating process. Common data mining tasks associated with data streams include clustering, classification and frequent pattern mining. New algorithms for these types of data are proposed regularly and it is important to evaluate them thoroughly under standardized conditions. In this paper we introduce stream, a research tool that includes modeling and simulating data streams as well as an extensible framework for implementing, interfacing and experimenting with algorithms for various data stream mining tasks. The main advantage of stream is that it seamlessly integrates with the large existing infrastructure provided by R. In addition to data handling, plotting and easy scripting capabilities, R also provides many existing algorithms and enables users to interface code written in many programming languages popular among data mining researchers (e.g., C/C++, Java and Python). In this paper we describe the architecture of stream and focus on its use for data stream clustering research. stream was implemented with extensibility in mind and will be extended in the future to cover additional data stream mining tasks like classification and frequent pattern mining.

Fat node leading tree for data stream clustering with density peaks

Detecting clusters of arbitrary shape and constantly delivering the results for newly arrived items are two critical challenges in the study of data stream clustering. However, the existing clustering methods could not deal with these two problems simultaneously. In this paper, we employ the density peaks based clustering (DPClust) algorithm to construct a leading tree (LT) and further transform it into a fat node leading tree (FNLT) in a granular computing way. FNLT is a novel interpretable synopsis of the current state of data stream for clustering. New incoming data is blended into the evolving FNLT structure quickly, and thus the clustering result of the incoming data can be delivered on the fly. During the interval between the delivery of the clustering results and the arrival of new data, the FNLT with blended data is granulated as a new FNLT with a constant number of fat nodes. The FNLT of the current data stream is maintained in a real-time fashion by the Blending-Granulating-Fading mechanism. At the same time, the change points are detected using the partial order relation between each pair of the cluster centers and the martingale theory. Compared to several state-of-the-art clustering methods, the presented model shows promising accuracy and efficiency.

State-of-the-art on clustering data streams

Clustering is a key data mining task. This is the problem of partitioning a set of observations into clusters such that the intra-cluster observations are similar and the inter-cluster observations are dissimilar. The traditional set-up where a static dataset is available in its entirety for random access is not applicable as we do not have the entire dataset at the launch of the learning, the data continue to arrive at a rapid rate, we can not access the data randomly, and we can make only one or at most a small number of passes on the data in order to generate the clustering results. These types of data are referred to as data streams. The data stream clustering problem requires a process capable of partitioning observations continuously while taking into account restrictions of memory and time. In the literature of data stream clustering methods, a large number of algorithms use a two-phase scheme which consists of an online component that processes data stream points and produces summary statistics, and an offline component that uses the summary data to generate the clusters. An alternative class is capable of generating the final clusters without the need of an offline phase. This paper presents a comprehensive survey of the data stream clustering methods and an overview of the most well-known streaming platforms which implement clustering.

A clustering algorithm for stream data with LDA-based unsupervised localized dimension reduction

We present an algorithm for clustering high dimensional streaming data. The algorithm incorporates dimension reduction into the stream clustering framework. When a new datum arrives, the algorithm performs dimension reduction to find a local projected subspace using unsupervised LDA (Linear Discriminant Analysis)-based method. The obtained local subspace would maximally separate the nearby micro-clusters with respect to the incoming point. Then, the incoming point is assigned to a micro-cluster in the projected space, rather than in the full dimensional space. The experimental results show that the proposed algorithm outperforms its counterpart streaming clustering algorithms. Moreover, when compared with traditional clustering algorithms which require the whole data set, the proposed algorithms shows comparable clustering performances with much less computation time for large data sets.

Speaker diarization system using MKMFCC parameterization and WLI-fuzzy clustering

Speaker diarization is the process of determining "who speak when?" with appropriate speaker labels with respect to the time regions where they spoke. Accordingly, in the previous work, a model based speaker diarization using the tangential weighted Mel frequency cepstral coefficients as the feature parameter for the voice activity detection and Lion optimization algorithm for the clustering of the audio streams into speaker group was performed. In this paper, speaker diarization system is proposed using multiple kernel weighted Mel frequency cepstral coefficient (MKMFCC) parameterization and Wu-and-Li Index (WLI)-fuzzy clustering. First, a MKMFCC which utilizes the multiple kernels like the tangential and exponential for weighting the MFCC's is proposed for the feature parameterization. Second, a clustering algorithm called the WLI-Fuzzy clustering is proposed for grouping the segments of the same speaker groups. The experimentation of the proposed speaker diarization system is carried out over the publically available ELSDSR corpus data set having the audio signal with seven different speakers. The performance evaluation of the proposed speaker diarization system is analysed using the measures such as diarization error rate, F-measure and false alarm rate. The results show that the proposed speaker diarization system proved better for tracking the active speakers from multiple speakers with improved tracking accuracy.

Missing data imputation for paired stream and air temperature sensor data

Stream water temperature is an important factor in determining the impact of climate change on hydrologic systems. Near continuous monitoring of air and stream temperatures over large spatial scales is possible due to inexpensive temperature recorders. However, missing water temperature data commonly occur due to the failure or loss of equipment. Missing data creates difficulties in modeling relationships between air and stream water temperatures. It also imposes challenges if the objective is an analysis, for example, clustering streams in terms of the effect of changes in water temperature. In this work, we propose to use a novel spatial–temporal varying coefficient model to impute missing water temperatures. Modeling the relationship between air and water temperature over time and space increases the effectiveness of imputing the missing water temperatures. A parameter estimation method is developed, which utilizes the temporal covariation in the relationship, borrows strength from neighboring stream sites, and is useful for imputing sequences of missing data. A simulation study is conducted to examine the performance of the proposed method in comparison with several existing imputation methods. The proposed method is applied to cluster streams with missing water temperatures into groups from 156 streams with meaningful interpretations.

Using internal evaluation measures to validate the quality of diverse stream clustering algorithms

Measuring the quality of a clustering algorithm has shown to be as important as the algorithm itself. It is a crucial part of choosing the clustering algorithm that performs best for an input data. Streaming input data have many features that make them much more challenging than static ones. They are endless, varying and emerging with high speeds. This raised new challenges for the clustering algorithms as well as for their evaluation measures. Up till now, external evaluation measures were exclusively used for validating stream clustering algorithms. While external validation requires a ground truth which is not provided in most applications, particularly in the streaming case, internal clustering validation is efficient and realistic. In this article, we analyze the properties and performances of eleven internal clustering measures. In particular, we apply these measures to carefully synthesized stream scenarios to reveal how they react to clusterings on evolving data streams using both k-means-based and density-based clustering algorithms. A series of experimental results show that different from the case with static data, the Calinski-Harabasz index performs the best in coping with common aspects and errors of stream clustering for k-means-based algorithms, while the revised validity index performs the best for density-based ones.

Supervised Adaptive Incremental Clustering for data stream of chunks

Many supervised clustering algorithms have been developed to find the optimal clusters for static datasets by presetting some parameters, but they are seldom suitable for dynamic datasets, such as the data stream of chunks. To find the optimal clusters of the data stream of chunks, a novel Supervised Adaptive Incremental Clustering (SAIC) algorithm is proposed. SAIC can cluster dynamic datasets of arbitrary shapes and sizes automatically. It includes learning and post-processing phases. In the learning phase, each cluster updates adaptively according to its learning rate that is calculated from its counter value. All data points are shuffled at each iteration in order to make SAIC insensitive to the input order of data points. In the post-processing phase, the outliers or boundary points are eliminated according to the counter value of each cluster and the number of iterations. Four synthetic datasets and fourteen UCI datasets are used to evaluate the performance of SAIC, respectively. The experiments on UCI datasets show that SAIC reaches to or outperforms some other supervised clustering algorithms and several unsupervised incremental clustering algorithms. In addition, three data stream of chunks are used to evaluate SAIC from different aspects, which shows SAIC has the scalability and incremental learning ability for the clustering of data streams of chunks.

An evolutionary algorithm for clustering data streams with a variable number of clusters

Several algorithms for clustering data streams based on k-Means have been proposed in the literature. However, most of them assume that the number of clusters, k, is known a priori by the user and can be kept fixed throughout the data analysis process. Besides the difficulty in choosing k, data stream clustering imposes several challenges to be addressed, such as addressing non-stationary, unbounded data that arrive in an online fashion. In this paper, we propose a Fast Evolutionary Algorithm for Clustering data streams (FEAC-Stream) that allows estimating k automatically from data in an online fashion. FEAC-Stream uses the Page–Hinkley Test to detect eventual degradation in the quality of the induced clusters, thereby triggering an evolutionary algorithm that re-estimates k accordingly. FEAC-Stream relies on the assumption that clusters of (partially unknown) data can provide useful information about the dynamics of the data stream. We illustrate the potential of FEAC-Stream in a set of experiments using both synthetic and real-world data streams, comparing it to four related algorithms, namely: CluStream-OMRk, CluStream-BkM, StreamKM++-OMRk and StreamKM++-BkM. The obtained results show that FEAC-Stream provides good data partitions and that it can detect, and accordingly react to, data changes.

Gaps in globular cluster streams: giant molecular clouds can cause them too

Abstract As a result of their internal dynamical coherence, thin stellar streams formed by disrupting globular clusters (GCs) can act as detectors of dark matter (DM) substructure in the Galactic halo. Perturbations induced by close flybys amplify into detectable density gaps, providing a probe both of the abundance and of the masses of DM subhaloes. Here, we use N-body simulations to show that the Galactic population of giant molecular clouds (GMCs) can also produce gaps (and clumps) in GC streams, and so may confuse the detection of DM subhaloes. We explore the cases of streams analogous to the observed Palomar 5 and GD1 systems, quantifying the expected incidence of structure caused by GMC perturbations. Deep observations should detect such disturbances regardless of the substructure content of the Milky Way's halo. Detailed modelling will be needed to demonstrate that any detected gaps or clumps were produced by DM subhaloes rather than by molecular clouds.

Clustering Big Data streams: recent challenges and contributions

Abstract Traditional clustering algorithms merely considered static data. Today's various applications and research issues in big data mining have however to deal with continuous, possibly infinite streams of data, arriving at high velocity. Web traffic data, surveillance data, sensor measurements and stock trading are only some examples of these daily-increasing applications. Since the growth of data volumes is accompanied by a similar raise in their dimensionalities, clusters cannot be expected to completely appear when considering all attributes together. Subspace clustering is a general approach that solved that issue by automatically finding the hidden clusters within different subsets of the attributes rather than considering all attributes together. In this article, novel methods for an efficient subspace clustering of high-dimensional big data streams are presented. Approaches that efficiently combine the anytime clustering concept with the stream subspace clustering paradigm are discussed. Additionally, efficient and adaptive density-based clustering algorithms are presented for high-dimensional data streams. Novel open-source assessment framework and evaluation measures are additionally presented for subspace stream clustering.

Distributed stream clustering using micro-clusters on Apache Storm

The recent need to extract real-time insights from data has driven the need for machine learning algorithms that can operate on data streams. Given the current extreme rates of data generation (around 5000 messages per second), these algorithms need to be able to handle data streams of very high velocity. Many current algorithms do not reach this requirement, in some cases processing only tens of messages per second.In this work we address the problem of limited achievable throughput of stream clustering by developing scalable distributed algorithms based on the micro-clustering paradigm that run on cloud platforms. We present two distributed architectures to execute the algorithms in parallel and implement these architectures on the Apache Storm stream processing platform. We demonstrate that we are able to gain close to an order of magnitude of improvement of performance in our experiments.

An Ensemble of Adaptive Neuro-Fuzzy Kohonen Networks for Online Data Stream Fuzzy Clustering

A new approach to data stream clustering with the help of an ensemble of adaptive neuro-fuzzy systems is proposed. The proposed ensemble is formed with adaptive neuro-fuzzy self-organizing Kohonen maps in a parallel processing mode. Their learning procedure is carried out with different parameters that define a nature of cluster borders' blurriness. Clusters' quality is estimated in an online mode with the help of a modified partition coefficient which is calculated in a recurrent form. A final result is chosen by the best neuro-fuzzy self-organizing Kohonen map.

Efficiency of Stream Processing Engines for Processing BIGDATA Streams

Background/Objectives: In this paper, there are real-time stream & big data stream applications running on a large amount of clusters & live migration of data using intensive & interactive tasks. Methods/Statistical Analysis: Traditionally, there are many stream processing engines which are going to manage the processing data in this streaming scenario, ongoing stream processing the data must be handle due to faults & stragglers within a second-scale latency. At present, there are efficient stream processing engines which can avoid these faults & stragglers by using the stream processing engines like APACHE SPARK & APACHE FLINK. Findings: In the real-time data stream processing there are some problem occurrences while processing the data, those are data latency, fault-tolerance, mutable data-sets and more stragglers. To avoid all these issues there is an efficient stream processing engines with additional added mechanisms like dolly retreat mechanism for avoiding stragglers and process data efficiently. Application/Improvements: Now a day’s demand for stream processing is increasing a lot which may be real-time streaming or normal cluster data transformation. Data has to be processed fast, so that companies are integrating these stream processing engines to changing their business conditions effectively in realtime for analyze their data more effectively with help of Fast Stream Processing Engines.

Hyper-cylindrical micro-clustering for streaming data with unscheduled data removals

We present a streaming data clustering algorithm which allows removals of data records at any arbitrary time. Unlike other existing algorithms whose objective is to track evolving clusters by letting the weight of old data decline with time, we consider the case when all data records do not have pre-specified lifetime which is the characteristic of many real data sets such as bank accounts. A data record can be added or removed by users at any arbitrary time. The algorithm processes each datum in one-pass-throw-away fashion without storing the whole data set. A technique for merging several micro-clusters into a hyper-cylindrical micro-cluster is proposed to reduce the number of micro-clusters in feature space, and thus reduce computation. The performance of this algorithm is tested with several data sets including both synthetic and real data sets. The proposed algorithm shows better performances compared with other state-of-the-art algorithms in terms of several indices for measuring clustering performance.

Data stream clustering by divide and conquer approach based on vector model

Recently, many researchers have focused on data stream processing as an efficient method for extracting knowledge from big data. Data stream clustering is an unsupervised approach that is employed for huge data. The continuous effort on data stream clustering method has one common goal which is to achieve an accurate clustering algorithm. However, there are some issues that are overlooked by the previous works in proposing data stream clustering solutions; (1) clustering dataset including big segments of repetitive data, (2) monitoring clustering structure for ordinal data streams and (3) determining important parameters such as k number of exact clusters in stream of data. In this paper, DCSTREAM method is proposed with regard to the mentioned issues to cluster big datasets using the vector model and k-Means divide and conquer approach. Experimental results show that DCSTREAM can achieve superior quality and performance as compare to STREAM and ConStream methods for abrupt and gradual real world datasets. Results show that the usage of batch processing in DCSTREAM and ConStream is time consuming compared to STREAM but it avoids further analysis for detecting outliers and novel micro-clusters.

Penalty Parameter Selection for Hierarchical Data Stream Clustering

Extracting useful information from large sets of data is the main task of data mining. Clustering is one of the most commonly used data mining technique. Data streams are sequences of data elements continuously generated at high rate from various sources. Data streams are everywhere and are generated by the applications like cell-phones, cars, security sensors, televisions and so on. Partitioning data streams into sets of meaningful subclasses is required for proper and efficient mining of intended data. Identifying the number of clusters required for the precise clustering of data streams is an open research area. This paper gives the overview of the hierarchical data stream clustering algorithms. It also compares the performance analysis of the different algorithms under hierarchical clustering techniques for data streams. Different data clustering tools are also explained and compared in this paper. It also applies the proper hierarchical clustering algorithm to the standard datasets taken as input and the expected result must be the clustered data which is well versed, properly arranged. This paper addresses the issue of identifying the number of clusters by proposed penalty parameter selection approach. The approaches presented in this paper are helpful for the researchers in the field of data stream clustering and data mining.

Topic Evolutionary Tweet Stream Clustering Algorithm and TCV Rank Summarization

Twitter which receives over 400 million tweets per day has emerged as an invaluable source of news, blogs, opinions and more. Our proposed work consist three components tweet stream clustering to cluster tweet using kmeans cluster algorithm and second tweet cluster vector technique to generate rank summarization using greedy algorithm, therefore requires functionality which significantly differ from traditional summarization. In general, tweet summarization and third to detect and monitors the summary-based and volume based variation to produce timeline automatically from tweet stream. Implementing continuous tweet stream reducing a text document is however not a simple task, since a huge number of tweets are worthless, unrelated and raucous in nature, due to the social nature of tweeting. Further, tweets are strongly correlated with their posted instance and up-to-the-minute tweets tend to arrive at a very fast rate. Efficiency-tweet streams are always very big in level, hence the summarization algorithm should be greatly capable. Flexibility-it should provide tweet summaries of random moment durations. Topic evolution-it should routinely detect sub-topic changes and the moments that they happen.