MINING TECHNIQUES FOR STREAMING DATA

International Journal of Data Mining & Knowledge Management Process (IJDKP),
Vol.12, No.2, March 2022
DOI:10.5121/ijdkp.2022.12201 1
MINING TECHNIQUES FOR STREAMING DATA
ManaL Mansour and Manal Abdullah
Department of Information Systems, King Abdulaziz University, Jeddah, Saudi Arabia
ABSTRACT
The huge explosion in using real time technology leads to infinite flow of data which known as data
streams. The characteristics of streaming data require different techniques for processing due its volume,
velocity and volatility, beside issues related to the limited storage capabilities. Hence, this research
highlights the significant aspects to consider when building a framework for mining data streams. It
reviews the methods for data stream summarizing and creating synopsis, and the approaches of processing
these data synopses. The goal is to present a model for mining the streaming data which describes the main
phases of data stream manipulation.
KEYWORDS
Data Stream, Data Mining, Data Synopsis, Micro-clusters.
1. INTRODUCTION
The amounts of the existed data globally almost reached to 64 zettabytes by the end of 2020, and
the forecasts indicate that it will grow to 175 zettabytes by the end of 2025 [1].This is
encouraging the data mining to become an integral part in many sectors such as businesses,
financial, medical, manufacturing, media, and many others. It helps to extract knowledge that
improve several processes like decision making, goals defining, customer personalizing,
decreasing risk and fraud.
The great interest in data mining leads to produce new data mining techniques that could process
huge infinite data flows that come from different sources such as sensors, network traffic, call
center records, social media posts, ATM and credit card operations, web searches, etc. All these
sources produce continuous sequences of massive data that generated in high speed rate and
known as data stream [2].
A data stream consists of infinite sequence of elements that keep flowing with high speed rate.
Every element of a stream is a pair of (Xi,Ti), where Xi is a dimensional vector arrives at time
stamp Ti. The time stamp Identifies when the element occurred [3]. Mining such type of data
streams is a continuous process of extracting knowledge and valuable insights from infinite and
rapid data signals. The main goal is to predict the class or the value of the new elements using
machine learning techniques such as clustering, classification, regression, etc [2].
Data stream is much related to the Big Data concept, since the big data comes as result of
streaming the data. The continuous flow of fast data causes too large volume of data which is
known as big data. It is important to clarify that streaming data growth and produce big data, but
it is not necessary that all big data comes from data stream. Big data usually reflects data at rest
mode even if it was petabytes while streaming data reflects data that in motion mode which need
a different technology to handle it [3].

2
Data mining techniques work differently with streaming data because of their different
characteristics such as data volume which keep increasing continually, data volatility which
reflects the different types of data values, and data velocity that reach very high ranges.
Therefore, the system requires specific methods to summarize, storing, and processing the data
stream [2]. Table 1 shows a comparison of mining traditional and streaming data [2].
Table 1. Comparison of Mining Traditional and Streaming data
Comparison Aspects Traditional Data Streaming Data
Data vitality Static data Dynamic data
Data size Fixed size Unlimited size
Number of data access Multiple access Single~few access
Processing procedure
Batch processing:
Process all points in the data set
Online Processing: Process
samples of data stream
Processing time Unlimited time Restricted time
Memory usage Unlimited storage Restricted storage
Results accuracy Accurate results Approximated results
2. DATA SYNOPSIS
Data streams can be generated using different types of artificial generators. Artificial generatoris
an embedded component in the mining model used to create a sequence of data simulating the
properties of the real data streams such speed, noise, density, etc. There are many existed data
streams generators such as: Random RBF Generator [4], SEA Concepts Generator [4],
AGRAWAL Generator [5], LED Generator [4], Hyperplane Generator [4], and Random Tree
Generator [5].
The characteristics of streaming data make it difficult to store all the data in a limited memory.
Instead, mining frameworks construct synopsis as a summary of the data stream. These synopses
can be constructed in different ways: randomly, by average, by trends, by most recent, etc.
Constructing the synopsis is a periodic online process which selects a subgroup of the data from
the running stream. Despite that using synopsis decreases the amount of processed data and the
analysis costs, it increases the possible errors. This problem makes it critical to choose the
suitable method for constructing the synopsis [6]. Next sections will review some of the main
methods of constructing the synopsis from the data stream.
2.1. Sampling
Sampling is one of the simplest methods for constructing synopsis from data streams. The
process of sampling the data stream depends on taking a subgroup of data elements that estimate
the whole statistical factors for streaming data, like mean, variance, probability distribution,
etc[2]. Figure 1 shows an example of sampling the streaming data with a simple rule of choosing
the data every four received data points.
Figure 1. Data Stream Sampling

3
There are many sampling methods and the challenge is to choose the one which produces a
sample that reflect the statistics of the original data. Random sampling is the simplest form,
where all data instances have equal probability to be included in the sample. One technique for
random sampling is the reservoir algorithm [7] which randomly collects data sample of size k and
keep updating this sample. Every new instance in the data flow has a probability p=1/n to replace
an old instance, where n is the number of elements that been generated from the stream. Other
sampling techniques include systematic sampling [8], stratified sampling [9], and cluster
sampling [10].
2.2. Windowing
Data stream windowing is a process of segmenting the data stream into small packets at certain
time t with specific size w [2]. Windowing approach helps to deals with concept drift which
reflects the difference in the data stream nature through time. The content of the window
continues to change and gives updated results whereas the element that arrives at time t is
rejected at time t+w [11]. There are four main types of time windows [12]: damped window,
sliding window, landmark window, and tilted window (pyramidal). Figure 2 describe the
windows types and how it works. All windows commonly update the data periodically, but they
differ in the updating procedure as described next sections.
Figure 2. Types of Data Stream Windowing [11]
2.2.1. Damped Window
This type of windows gives specific weight value for each arrived data element based on time.
New data elements have higher weight values which keep decreasing over time based on ageing
function. Ageing function used to update the window by using fading factor (λ) which controls
the rate of weight decreasing. All the data are involved in the processing, but with different
weight values which means different importance levels. Equation 1 presents the exponential
fading function, where the current time is denoted by tc and the point creation time is denoted by
t0 [12].
f(t) = 2λ (tc-
t0)
(1)
2.2.2. Sliding Window
This type of windows is more suitable for applications that interested in the newest information
because it considers only the recent data of the stream [12]. The window works based on the
principle (First-In-First-Out), where the old data are removed to add the new data. All the data
elements in the updated window have the same weight in processing which mean having the
same importance. The small size windows give better accuracy in high changing environment
with risks of data drifts, while the larger size works better in the stable data environment [11].

4
2.2.3. Landmark Window
This type of windows works by segmenting the data stream into chunks based on landmarks.
Landmark could reflect the data events or the passed time. For example, the landmark defined as
the starting point that could be (every 100 elements) which means updating the window after
receiving 100 new data points and removed all the previous 100 data points. The window
includes a summary of all data points that came after the landmark and gives them the same level
of importance [11] [12].
2.2.4. Pyramidal Window
This type of windows also known as tilted windows and it includes data from different times with
different detailing levels. The new coming data points are included to the window and aggregated
with the points received before. In contrast of sliding window, titled window does not completely
discard the old data, but integrate all the data and give more importance to the recent data [11]
[12].
2.3. Histograms
Histogram is a popular method for constructing synopsis of data stream which depend on
capturing the frequency distribution of the data values received from the stream. Instead of store
the frequency for each data value, the domain of data values are divided into subsets called
buckets. Then, the average frequency is computed and stored for each bucket [2].
Different types of data stream histograms are developed with different manners of dividing the
data domain into buckets. The histogram with less variance is better, but it requires more space
capabilities to implants. The histogram variance defined as the weighted sum of the original
values for each bucket where the weight of each bucket equals the number of its included values
[13] [14]. There are five types of histograms used to store data distributions which presented in
figure 3.

5
Figure 3. Types of Data stream Histograms [14]
2.3.1. Equi-width histograms
This type works by dividing the data domain into equal width buckets which means that every
bucket has the same range of data pints numbers. It reduces the storage requirements and works
well with simple skewed data distribution, but it is difficult to estimate the errors [13].
2.3.2. Equi-depth Histograms
This type works by dividing the data domain into similar height buckets which mean that data
points in certain bucket have almost the same frequencies. It requires the same storage of Equi-
width histograms, but it is more complex to maintain. It Works well with low skewed data
distribution and gives better accuracy than Equi-width histograms, but variance within a bucket
may still high [13].
2.3.3. Variance-optimal Histograms
This type works by dividing the data domain into buckets where the data frequencies in every
bucket is either all greater than or all less than the frequencies of other buckets. It works well
with high skewed data distributions and gives the minimum histogram variance, but it requires
high memory capacities [14].
2.3.4. MaxDiff Histograms
This type works by dividing the data points with the highest frequencies and lowest frequencies
in individual separated buckets, while the data with middle frequencies are all combined with one

6
bucket. It stores only the frequencies of the attribute values in the individual buckets which
require smaller storage capacities than Variance-optimal histograms [14].
4.3.5. Compressed Histograms
This type works by dividing the data points with high frequencies in singleton buckets while the
rest data is divided as equi-depth histogram. It gives great accuracy with skewed data
distributions [14].
2.4. Sketching
The data sketching is a process of creating a structure of the data stream. The data sketch includes
some information about data samples that help to response some predefined questions [15]. The
stored information could be the number of the received values appears in the stream without
repetition like introduced in Flajolet Martin Algorithm (FM Sketch) [16], the most frequent
elements seen in the stream like introduced in Count Sketch Algorithm [17], or frequencies of
data events like introduced in Count–Min Algorithm (CM Sketch) [18]. The task of recording
values frequencies is shared in CM sketching and histograms techniques discussed before, but
histogram considers the frequencies of specific data sample, while CM sketching updates the
frequencies of a data sample over the stream [15]. Figure 4 shows an example of count sketching
in top view.
Figure 4. Count Sketching for Data Stream
Sketching is much related to sampling technique since it uses some initials samples as basis to
build the sketch. Every coming data point will be discarded after being used to update the sketch
which means not storing all the data dimensions, but only the interested dimensions. This
technique presents the sketching as a powerful method to compress data stream with high
dimensions [15].
3. DATA STREAM PROCESSING
Data stream processing is a continuous operational techniques applied on the live data to give
direct results [3]. Figure 5 shows the basic approaches of processing data stream: supervised
learning, unsupervised learning. The main difference is about using labelled data or not. All
approaches that depend on labelled data are assumed as supervised where each input is assigned
to a predefined label based on knowledge comes from training data examples. In contrast, the
approaches that depend on unlabelled data are assumed as unsupervised where the main tasks are

7
about finding the similarities between data points. Some approaches assumed as semi supervised
since it uses both types: some labelled data at the early stages and then continue the process with
unlabeled data [19].
Figure 5. Machine Learning Approaches
3.1. Classification
Classification is a process of data organizing and grouping based on predefined classes which
also called labels, For example, predicting the type of fruit. The process consists of two phases:
training and prediction which are clearly separated in case of batch processing for regular and
static data. In training phase, the dataset is divided into two sets: training and testing to ensure
the accuracy. Then, the prediction phase starts the actual classification process.
In contrast of static data, classifying streaming data doesn't have a clear cut between training and
prediction phases. Training, evaluating, and testing are overlapped since more data is arriving
continuously and the prediction phase starts before receiving all data. The algorithm predicts the
labels of the received data based on prediction model and continues to update and adjust that
model while receiving new data [3]. Figure 6 shows how some approaches update the model
with every single received data instance, while some others update the model with every block of
data points [20].
Machine Learning
Approaches
Supervised
Classification
Regression
Unsupervised
Clustering
Assosiation

8
Figure 6. Updating Prediction Model of Data Stream [20]
Several classifiers were developed to solve the problem of predicting the data stream such as:
Lazy classifier [21]. Decision Trees [22]which works by building a tree structure using training
examples that used to predict the labels of other coming examples, and Naïve Bays [23] which is
a simple classifier works based on the use of the Bayes’ theorem.
3.2. Regression
The regression classifier aims for predicting the label value of an example where data labels is
not yet known, and then classify the data to the closest label. The predicted label is a real value
instead of being one of the discrete set of predefined values, For example, predicting the house
price which will assign each house to the closet price label. Some classification algorithms
include natural counterparts for regression, such as: lazy learning and decision trees [3].
3.3. Clustering
Clustering is a process of finding homogeneous groups or clusters among unlabeled data. It is a
useful technique in some problems such as segmenting the customers to find the targeted groups,
and finding communities in social networks [2].
Clustering approaches for data streams consist of two steps: live data phase and offline data
phase. In the live data phase, the algorithm performs one pass over some current data streams to
computes a specific number of micro-clusters, store it in memory, and update it using new
subsets of the coming new data. In the offline phase, the algorithm has several passes only over
the stored micro-clusters to processes them using traditional batch clustering techniques [3]. Next
subsections will describe some of the common algorithms for clustering the streaming data.

9
3.3.1. Hierarchical Clustering
This approach works by grouping the data into a tree of clusters such as Clustree Algorithm [24],
and Brich Algorithm [25].
3.3.2. Partitioning Clustering
This approach works by grouping the data into number of partitions which reflects the clusters. It
works based on distance functions like k-means which used in StreamKM++ Algorithm [26].
3.3.3. Density Based Clustering
This approach is based on density functions which assume dense areas of data points as clusters.
It requires two values for the cluster: the radius size and the minimal number of data points to
include in the cluster. DenSteam Algorithm [27] is a popular example of this approach which
works based on summarizing clusters using dense micro-clusters called core-micro-cluster.
3.3.4. Grid Based Clustering
This approach is based on computing the grid density. D-Stream Algorithm [28] is an example
which works by segmenting the data space into finite number of cells assumed as grid structure.
Then, mapping the recorded data to that grid which will be clustered based on their density.
3.3.5. Growing Neural Gas Clustering
The growing neural gas is a clustering approach works by devising the data space into units
depending on the densest areas. It produces a two-dimensional presentation for the clusters and
the data patterns [29].
3.4. Association
Finding association rules is a process of finding frequent pattern among data such as trees,
graphs, item sets, etc. It is useful to describe the structure of the data and explore its features
which could enhance other processes of classification or clustering. Most algorithms of finding
frequent items for data streams includes batch miner and streaming techniques which produce
approximate results [2]. IncMine algorithm [30] is one of the algorithms developed to find the
frequent items among streaming data. It works by updating the model incrementally over a
sliding window.
4. REAL APPLICATIONS AND TOOLS
The importance of data stream analytics increased as a result of turning most organizations and
business towards digital work and applications which produce real time data. This is encourages
to develop mining tools to help get benefits from streaming data.
Table 2 shows several real applications of mining streaming data such as clinical decision support
system which helps in disease diagnosis, choosing medicines and improving the patient care [31].
Applications also include the biological field such as using the satellite systems to study the earth
nature. Moderate Resolution Imaging Spectroradiometer (MODIS) [32] is an example which
processes the satellite signals to study the long-term change of vegetation.

10
Many other applications related to the social fields are developed such as mining the data stream
of social media like Twitter, LinkedIn, Facebook, etc. All these types of platforms produce
unstructured data streams which needs machine learning techniques to analyze users profiles,
identify posts patterns, trends, and others insights. Novel metaheuristic method (CSK) proposed
in [33] is an example of social applications which developed for sentiment analysis of Twitter. It
works based on K-means method combined with cuckoo search technique to provide reviews
conclusion for any social issues.
Marketing is one of the biggest beneficiaries’ fields which use data stream analysis. Most markets
are interested in studying the customers’ behavior which could be done by segmenting the
customers into groups based on their purchases baskets. Studying the customers’ behavior helps
to customize special offers for them. Sales markets also interested to study the relationships
between different product items which help in designing the items catalog, deciding the profitable
offers [3]. Other markets applications include Stock market prediction. One example of the
applications used in on this field is using Neural Network and Decision Tree which proposed in
[34]. This algorithm use data of markets sales, and the customers’ interests to predict the future
value of the companies’ stock.
Table 2. Real Applications of Data Stream Mining
Application field Streaming data Mining Objective
Medical [31] Health status records Disease diagnosis, and improving the patient care
Biological [32] Satellite signals Study the long-term change of vegetation
Social [33] Twitter posts Provide reviews conclusion for any social issues
Marketing [34] Sales data Prediction of Stock Market
Many frameworks appear to process data streams which make it critical to choose the suitable
one that meets the problem requirements. Waikato Environment for Knowledge Analysis
(WEKA) [35], Massive Online Analysis (MOA) [3], Hadoop [36], Spark [37], Storm [38], and
Flink [39] are the some popular processing frameworks for mining data streams.
Existing tools for mining data stream differ in their models wither it works with batch, real time
processing, or both. Batch processing collects the data over specific period of time like payroll
and billing data. In contrast, real time processing receives continual data like bank services and
social media messages. Table 3 reviews the processing model, programing languages, and main
functions of some existing tools for mining the data stream.
.
Table 3. Tools for Mining Data Stream
Tool Processing
Model
Program
language
Functions
WEKA
[35]
Hybrid:
batch and real-
time
Java-Python
Includes a large set of machine learning algorithms
used in data mining operations
MOA
[3]
Real time only java
includes a several types of data streams generators,
algorithms of machine learning , and tools for
results evaluation
Hadoop
[36]
Batch only
Java-Python
and Scala
Provides distributed storage for big data processing
based on MapReduce programming model.

11
Spark
[37]
Hybrid:
batch and real-
time
Java-Python
and
Scala
General-purpose processing engine for both big
and steaming data
Storm
[38]
Real time only
Any
programming
language
Distributed framework for stream processing
computation. Very fast and capable to integrate
with Hadoop for immediate analytics
Flink
[39]
Hybrid:
batch and real-
time
Java, Scala,
Python and
SQL
Distributed processing engine and a scalable data
analytics framework. includes data stream API
5. RELATED WORK
The data stream analysis is a rich field and always required for further improvements. Several
contributions were adopted to create, develop, and improve the algorithms of mining data stream.
The contributions in data stream algorithms may be through changing the initialization methods
such proposed in [40], changing the method of constructing the synopsis of data stream or
merging two methods such suggested in [41], integrating multiple methods of data stream
processing as proposed in [42], or implementing the algorithm in a new field such produced [43].
Other related work and contributions were reviewed in table 4.
Table 4. Algorithms for Data Stream Mining
Algorithm Synopses Type Processing
Approach
Main Ideas
Very Fast Decision Tree
(VFDT) [31]
Samples Classification Using The VFDT with pointers to
enhance the decision tree
Hyper-Ellipsoidal
Clustering for Evolving
data Stream (HECES)
[40]
Sliding window Clustering Using covariance shrinkage for
data estimation, and heuristic
technique for the initial clusters
A Privacy-Preserving
Distinct Counting
Scheme for Mobile
Sensing (PPDC) [41]
FM Sketches Clustering Integrating homomorphic data
encrypting with Flajolet-Martin
sketching
Scale-free social network
(SFNClassifier) [42]
Sliding window Classification Integrating Adwin Algorithm
with ensemble-based approaches
Clustering Algorithm for
geographical data
streams (GeoDenStream)
[43]
Damped window Clustering Enhancing DenStream Algorithm
to apply entity-based clustering
on geographical spaces
Uncertain Data Stream
Frequent Itemsets Mining
(UFIM) [44]
Sliding window Association Using a global GT-tree to process
and update the frequent items in
the sliding windows
Adaptive Random Forest
Algorithm (ARF) [45]
Samples Classification Implementing resampling method
for detecting data drifts and other
techniques like warning detection
and background trees
Parsimonious Learning
Machine Algorithm
(PALM) [46]
Samples Regression Using upgraded rule of fuzzy
method which developed based
on the idea of hyperplane
clustering
Augmented Sketch
Algorithm (ASketch)
[47]
Sketches Association Improving the Count-Min sketch
by implementing a pre-filtering
stage for data stream

12
6. CONCLUSION AND FINDINGS
Today's digital world emphasizes the important role of data streams analytics which helps to
develop the operations and improve the performance in many fields. Despite the great evolution
in data mining techniques, it still have difficulties to deal with the huge explosion of data comes
from real time applications. Many applications were developed for mining the streaming data,
and there are common phases in their architectures which are: data generating, sampling,
initializing, and processing. These phases combined to define the general data streams model
presented in Figure 7.
Figure 7. General Model for mining data streams
The general model presents the process of data streams manipulation following the common steps
of the existing models. First, streaming data flows to the system continuously in massive
volumes and high velocity. Since the system is unable to process the whole stream, it takes some
initial data, processes it, produce approximated results as a summary, and store it in memory.
Then, the system takes different samples from the new coming data periodically, and updates the
initial results stored in memory. Finally, only the updated summaries will be processed to
introduce the final results instead of using the raw data comes from the stream.
These Three phases offer good flexibility to explore the evolution nature of the data over
different time periods which means handling the data streams over time rather than mining the
whole stream at one time.
7. RECOMMENDATIONS
The process of mining the data stream keeps developing to enhance the results quality. There are
some suggestions that are recommended to extend the work:
 Applying the model with real and artificial data generators
 Evaluating the model while raising the data stream speed and noise

13
 Enhancing the model by using other methods for data summarizing and creating
synopsis.
REFERENCES
[1] “How Much Data Is Created Every Day,” 28 October 2021. [Online]. Available:
https://seedscientific.com/how-much-data-is-created-every-day/. [Accessed 18 12 2021].
[2] M. Garofalakis, J. Gehrke and R. Rastogi, in Data Stream Management: Processing High-Speed Data
Streams, Springer, 2016.
[3] A. Bifet, R. Gavalda, G. Holmes and B. Pfahringer, Machine Learning for Data Streams:with
Practical Examples in MOA, MIT Press, 2018.
[4] P. K. SRIMANI and M. M. PATIL, “Mining data streams with concept drift in massive online
analysis frame work,” WSEAS Trans. Comput, vol. 15, pp. 133-142, 2016.
[5] O. Wu, Y. S. Koh, G. Dobbie and T. Lacombe, “Transfer Learning with Adaptive Online
TrAdaBoost for Data Streams,” in Asian Conference on Machine Learning, 2021.
[6] Z. Shah, A. N. Mahmood, Z. Tari and A. Zomaya, “A Technique for Efficient Query Estimation,”
IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS, vol. 28, pp. 2770-
2783, 2017.
[7] Gaber, Mohamed Medhat; Gama, Jo˜ao ; Krishnaswamy, Shonali; Gomes, Jo˜ao B´ artolo ; Stahl,
Frederic;, “Data stream mining in ubiquitous environments: state-of-the-art and current directions,”
WIREs: Data Mining & Knowledge Discovery, vol. 4, no. 2, pp. 116-138, 2014.
[8] D. J. Brus and N. Saby, “Approximating the variance of estimated means for systematic random
sampling, illustrated with data of the French Soil Monitoring Network,” Elsevier, vol. 279, pp. 77-86,
2016.
[9] S. S. Ramkrishna and . P. S. Housila, “Efficient classes of estimators in stratified random,” Statistical
Papers, vol. 56, no. 1, pp. 83-103, 2015.
[10] W. Li , “Joint Image-Text News Topic Detection and Tracking by Multimodal Topic And-Or Graph,”
IEEE transactions on multimedia, vol. 19, no. 2, pp. 367-381, 2017.
[11] M. Carnein and H. Trautmann , “Optimizing data stream representation: An extensive survey on
stream clustering algorithms,” Business & Information Systems Engineering, vol. 61, no. 3, p. 277–
297, 2019.
[12] E. Ntoutsi, N. Pelekis and Y. Theodoridis, “An evaluation of data stream clustering algorithms,”
Statistical Analysis and Data Mining, vol. 11, no. 4, pp. 167-187, 2018.
[13] Y. Ioannidis, “The history of histograms (abridged),” Proceedings, pp. 19-30, 2003.
[14] J. Gama and T. Mendonça, “Constructing fading histograms from data streams,” PROGRESS IN
ARTIFICIAL INTELLIGENCE, vol. 3, no. 1, pp. 15-28, 2014.
[15] R. Jayaram, “Sketching and Sampling Algorithms for,” Carnegie Mellon University Pittsburgh,
Pittsburgh, 2021.
[16] P. Flajolet and G. N. Martin, “Probabilistic counting algorithms for data base applications,” Journal
of computer and system sciences, vol. 31, no. 2, pp. 182-209, 1985.
[17] M. Charikar, K. Chen and M. Farach-Colton, “Finding frequent items in data streams,” International
Colloquium on Automata, Languages, and Programming, vol. 2380, pp. 693-703, 2002.
[18] D. Ting, “Count-Min: Optimal Estimation and Tight Error Bounds using Empirical Error
Distributions,” in Proceedings of the 24th ACM SIGKDD International Conference on Knowledge
Discovery & Data Mining, 2018.
[19] A. C. Gorgônio, A. . M. d. P. Canuto, K. M. O. Vale and F. . L. Gorgônio, “A semi-supervised based
framework for data stream classification in non-stationary environments,” International Joint
Conference on Neural Networks (IJCNN), pp. 1-8, 2020.
[20] D. Brzezinski and J. Stefanowski, “Stream Classification.,” in Encyclopedia of Machine Learning,
Springer, 2016, pp. 1191-1199.
[21] J. Yuan, Z. Wang, Y. Sun, W. Zhang and J. Jiang, “An effective pattern-based Bayesian classifier for
evolving data stream,” Neurocomputing, vol. 295, pp. 17-28, 2018.
[22] D. Jankowski, K. Jackowski and B. Cyganek, “Learning Decision Trees from Data Streams with
Concept Drift,” Procedia computer science, vol. 80, pp. 1682-1691, 2016.
[23] M. Vadivukarassi, N. Puviarasan and P. Aruna, “Sentimental Analysis of Tweets Using Naive Bayes
Algorithm,” World Applied Sciences Journal, vol. 35 (1), pp. 54-59, 2017.

14
[24] J. Zgraja and M. Woźniak, “Drifted data stream clustering based on ClusTree algorithm,”
International Conference on Hybrid Artificial Intelligence Systems, vol. 10870, pp. 338-349, 2018.
[25] G. Pitolli, L. Aniello, G. Laurenza, L. Querzoni and R. Baldoni, “Malware family identification with
BIRCH clustering,” International Carnahan Conference on Security Technology , pp. 1-6, 2017.
[26] M. R. Ackermann, M. Märtens, C. Raupach, K. Swierkot, C. Lammersen and C. Sohler,
“StreamKM++: A Clustering Algorithms for Data Streams,” Journal of Experimental Algorithmics
(JEA), vol. 17, pp. 2-1, 2012.
[27] F. Cao, M. Estert, W. Qian and A. Zhou, “Density-Based Clustering over an Evolving Data Stream
with Noise,” in Proceedings of the 2006 SIAM International Conference on Data Mining (SDM),
2006.
[28] M. H. Ali, A. Sundus, W. Qaiser, Z. Ahmed and Z. Halim, “Applicative implementation of D-stream
clustering algorithm for the real-time data of telecom sector,” nternational conference on computer
networks and information technology, pp. 293-297, 2011.
[29] M. Ghesmoune, M. Lebbah and H. Azzag, “A new Growing Neural Gas for clustering data streams,”
Neural Networks, vol. 78, pp. 36-50, 2016.
[30] J. Cheng, Y. Ke and W. Ng, “Maintaining frequent closed itemsets over a sliding window,” Journal
of Intelligent Information Systems, vol. 31, p. 191–215, 2007.
[31] Z. Yang , S. Fong, J. Fiaidhi and S. Mohammed, “Real-Time Clinical Decision Support System with
Data Stream Mining.,” Journal of Biomedicine & Biotechnology, pp. p1-8, 8p, 2012.
[32] J. J. Maynard, J. W. Karl and D. M. Browning, “Effect of spatial image support in detecting long-
term vegetation change from satellite time-series,” Landscape Ecology, vol. 31, no. 9, pp. p2045, 18
p, 2016.
[33] C. P. Avinash , “Twitter sentiment analysis using hybrid cuckoo search method,” Information
processing & management, vol. 53, no. 4, p. 764, 2017.
[34] S. Tiwari and A. Gulati, “Prediction of Stock Market from Stream Data Time Series Pattern using
Neural Network and Decision Tree,” Citeseer, 2011.
[35] E. Frank, . M. A. Hall and I. H, “Weka 3: Machine Learning Software in Java,” University of
Waikato, 2016. [Online]. Available: https://www.cs.waikato.ac.nz/ml/weka/. [Accessed 2022].
[36] A. Hadoop®, “What Is Apache Hadoop?,” hadoop.apache.org, 18 10 2018. [Online]. Available:
http://hadoop.apache.org/. [Accessed 2 12 2018].
[37] “Apache Spark™ - Lightning-Fast Cluster Computing,” The Apache Software Foundation, 2 3 2017.
[Online]. Available: https://spark.apache.org/. [Accessed 2 12 2018].
[38] “Apache Storm,” Apache Software Foundation, 3 6 2017. [Online]. Available:
http://storm.apache.org/. [Accessed 2 12 2018].
[39] “Apache Flink: Introduction to Apache Flink®,” The Apache Software Foundation, 7 2 2017.
[Online]. Available: https://flink.apache.org/introduction.html. [Accessed 2 12 2018].
[40] Z.-u. R. Muhammad, . L. Tianrui, Y. Yan and W. Hongjun , “Hyper-ellipsoidal clustering technique
for evolving data stream,” Knowledge-Based Systems, vol. 3, no. 14, pp. 3-14, 2014.
[41] X. Yang , M. Xu, S. Fu and Y. Luo , “PPDC: A Privacy-Preserving Distinct Counting Scheme for
Mobile Sensing,” Applied Sciences, vol. 9(18), p. 3695, 2019.
[42] J. P. Barddal, H. M. Gomes and F. Enembreck, “SFNClassifier: A scale-free social network method
to handle concept drift,” in Proceedings of the 29th Annual ACM Symposium on Applied
Computing, 2014.
[43] M. Li, A. Croitoru and S. Yue, “GeoDenStream: An improved DenStream clustering method for
managing entity data within geographical data streams,” Computers & Geosciences, vol. 144, p.
104563, 2020.
[44] H. Liu, . K. Zhou, P. Zhao and S. Yao, “Mining frequent itemsets over uncertain data streams,”
nternational Journal of High Performance Computing and Networking, vol. 11, pp. 312-321, 2018.
[45] H. . M. Gomes, A. Bifet, J. Read, J. P. Barddal, F. Enembreck and B. Pfharinger, “Adaptive random
forests for evolving data stream classification,” Machine Learning, vol. 106, pp. 1469-1495, 2017.
[46] M. M. Ferdaus, M. Pratama, S. . G. Anavatti and M. A. Garratt, “Palm: An incremental construction
of hyperplanes for data stream regression,” IEEE Transactions on Fuzzy Systems, vol. 27, no. 11, pp.
2115-2129., 2019.
[47] P. Roy, A. Khan and G. Alonso, “Augmented sketch: Faster and more accurate stream processing,” in
Proceedings of the 2016 International Conference on Management of Data, 2016.

MINING TECHNIQUES FOR STREAMING DATA

More Related Content

Similar to MINING TECHNIQUES FOR STREAMING DATA

Recently uploaded

MINING TECHNIQUES FOR STREAMING DATA