Improved K-mean Clustering Algorithm for Prediction Analysis using Classification Technique in Data Mining

Improved K-mean Clustering Algorithm for
Prediction Analysis using Classification
Technique in Data Mining
Swati Vinodani
Galgotia college of engineering
Swativinodani2015@gmail.com
Aatif jamshed
Assistant professor
Galgotia college of engineering
09.aatif@gmail.com
Dr. Pramod Kumar
Director
Tula`s Institute of Technology
drpramoderp@live.com
Abstract- Data mining is utilized to manage huge measure of information which are put in the data ware houses and
databases, to discover required information and data. Numerous data mining systems have been proposed, for example,
association rules, decision trees, neural systems, clustering, and so on. It has turned into the purpose of consideration from
numerous years. A re-known amongst the available data mining strategies is clustering of the dataset. It is the most effective
data mining method. It groups the dataset in number of clusters based on certain guidelines that are predefined. It is
dependable to discover the connection between the distinctive characteristics of data.
In k-mean clustering algorithm, the function is being selected on the basis of the relevancy of the function for
predicting the data and also the Euclidian distance between the centroid of any cluster and the data objects outside the
cluster is being computed for the clustering the data points. In this work, author enhanced the Euclidian distance formula
to increase the cluster quality.
The problem of accuracy and redundancy of the dissimilar points in the clusters remains in the improved k-means
for which new enhanced approach is been proposed which uses the similarity function for checking the similarity level of
the point before including it to the cluster.
Keywords: Data Mining, Clustering, Classification, Dataset, k-means, Similarity, Centroid, Data objects, Density.
International Journal of Computer Science and Information Security (IJCSIS),
Vol. 16, No. 5, May 2018
227 https://sites.google.com/site/ijcsis/
ISSN 1947-5500

I. INTRODUCTION
Data mining, the acquiring of required data from huge databases, is an intensive new development in the field with
high potential to enable organizations to center around the critical data in their dataset. Different methodologies of
data mining tries to extract the pattern on the basis of the future requirement from the dataset and do the job to make
the organization to have the active or proactive, learning based kind of work. The different works of the data mining
makes the organization to answer the issues related to the dataset which generally are tiring work and also sometime
are just impossible kind of work. Some of the concealed cases are being considered so as to scour the database and
makes the prescient data available which actually was forbidden by the works done previously [1].
Most organizations officially gather’s and refine huge amounts of information. Data mining procedures can
work quickly on existing tols and equipment to uplift the advantages of available data resources and can coordinate
with new items and frameworks as they are made on-line. At the point when runned on superior customer/server or
parallel preparing PCs, data mining devices can manage large databases to convey answers to inquiries, for example,
"Which customers are well on the way to react to my next promotion-based mailing, and why?"
Clustering is a data mining procedure that makes noteworthy or supportive group of substance that have
comparative element utilizing mechanical strategy. Divergent from order, Clustering procedure likewise characterizes
the classes and place data points in them, as in classification objects are considered into predefined classes. For
instance, in forecast of coronary illness by utilizing Clustering, get group or express that rundown of patients which
have same hazard factor. This influences the split rundown of patients with high blood to sugar and related hazard
factor n so on.
The idea of a "cluster" can't be accurately characterized, which is one reason whythere are such huge numbers
of clustering algorithms. [3] There is a shared factor: a groups of information objects. As, distinctive analysts utilizes
diverse cluster models, and for every one of these cluster models again extraordinary techniques can be given. The
thought of a cluster, as found by various techniques, shifts fundamentally in its properties.
A. ClusteringMethods
Clustering methods can be classified into the following categories:
 Partitioning Method
 Hierarchical Method
 Density-based Method
 Grid-Based Method
 Model-Based Method
 Constraint-based Method
Vol. 16, No. 5, May 2018
ISSN 1947-5500

B. Data Mining Used in Various Applications [2]
 Business Intelligence
 Sports
 Analyze Students Performance
 Telecommunication Industry
 Retail Industry
C. Issues and Challenges:
 Dissimilarity of the data in the clusters,
 Clustering accuracy,
 To produce fixed and appropriate cluster centers,
II. PROBLEM STATEMENT
In our work we extend the work done by Arpit et al. (2017) [4], “Improved K-mean Clustering Algorithm for
Prediction Analysis using Classification Technique in Data Mining”, k-means algorithm is being used for database
clustering in which the centroid is calculated and then on the basis of the Euclidian distance the data objects from the
dataset are then grouped to form of cluster of similar type of clusters. As the data object inclusion in the cluster is
being using the Euclidian distance hence in some of the cases the methodology fails to show the accuracy of the
clusters in the terms of similarity of the data objects in any clusters and also the major disadvantage of the k-means is
the static definition of the threshold for the number of clusters. The major disadvantage of k-means algorithm is that
the number of cluster for any dataset are needed to be predefine and just because of which some of the points or objects
inside the dataset remains un-clustered.
III. LITERATURE REVIEW
[4] The k-mean clustering techniques is utilized to group the comparative type of information for prediction analysis.
In k-mean clustering techniques, probability of the most relevant function is figured and utilizing Euclidian distance
equation the objects are grouped. In this work, we will improve the Euclidian distance equation to expand the cluster
quality. The upgrade will be based on normalization. In the improvement two new highlights will be included. The
primary point is to compute ordinary distance measurements based on normalization. In second point the objects will
be grouped based on larger part voting. The proposed strategy will be executed in MATLAB.
[5] K-Means or Hard C-Means clustering is essentially a dividing strategy connected to investigate
information and considers perceptions of the information as data point in view of areas and separation between
different information points. Partitioning the items into totally unrelated groups (K) is finished by it in such a way, to
the point that articles inside each cluster stay as close as possible to each other, and also far from objects in different
groups.
Vol. 16, No. 5, May 2018
ISSN 1947-5500

Each cluster is represented by its inside point i.e. centroid. The distance utilized as a part of clustering in the
vast majority of the circumstances don't really speak to the spatial separations. By and large, the main answer for the
issue of finding worldwide least is comprehensive decision of beginning stages. Yet, utilization of a few copies with
random beginning stage prompts an answer.
[6] Bezdek presented Fuzzy C-Means clustering strategy in 1981, stretch out from Hard C-Mean clustering
technique. FCM is an unsupervised clustering technique that is connected to extensive variety of issues associated
with feature investigation, clustering and classifier design. FCM is broadly connected in agrarian engineering,
chemistry, geography, image investigation, medical analysis, shape analysis and target acknowledgment [7].
With the improvement of the fuzzy hypothesis, the FCM clustering techniques which is really in relies on
Ruspini Fuzzy grouping hypothesis proposed in 1980's. This technique is utilized for analysis of distance between
different information objects. The clusters are shaped by the distance between information objects and the cluster
center are framed for each cluster.
[8] The DBSCAN techniques was first presented by Ester, et al. [Ester1996], and depends on a density-based
idea of clusters. Clusters are recognized by considering the thickness of data points. Areas with a high-density objects
shows the presence of clusters though locales with a low density of points demonstrate clusters of noise or clusters of
exceptions. This technique is especially suited to manage huge datasets, with noise, and can distinguish cluster with
various sizes and shapes.
The key thought of the DBSCAN techniques is that, for each object of a cluster, the area of a given radius
needs to contain no less than a base number of data points, that is, the density in the area needs to surpass some
predefined limit.
[9] The SNN technique [Ertoz2003], as DBSCAN, is a thickness-based clustering technique. The
fundamental distinction between this technique and DBSCAN is that it characterizes the comparability between points
by considering at the quantity of closest neighbors that two points share. Utilizing this similarity measure in the SNN
technique, the density is characterized as the aggregate of the similarity of the closest neighbors of a point. Points with
high density move toward becoming center points, while points with low density speak to noise points. All leftover
points that are emphatically comparative a particular center points will speak to another clusters.
[10] DENCLUE (Density based clustering) utilizes two primaryideas i.e. impact and density function. Impact
of every data point can be displayed as numerical function. The subsequent function is called Influence Function.
Impact work defines the effect of information point inside its neighborhood. Second factor is Density function which
is total of impact of all information points. DENCLUE characterizes two kinds of clusters i.e. defined characterized
and multi center characterized groups. y Є F is an impact function of the information objects.
Which is characterized as far as an essential impact function F, F (x) =-F (x, y). The density function might
be characterized as the whole of the impact elements of all information points. DENCLUE is additionally used to sum
Vol. 16, No. 5, May 2018
ISSN 1947-5500

up other clustering techniques like Density based grouping, segment-based clustering, hierarchical clustering.
DBSCAN is a case of density based clustering and square wave impact function is utilized.
[11] Fundamentally, DBCLASD is an incremental approach. DBCLASD depends on the assumption that the
points inside a cluster are grouped consistently. DBCLASD progressively decides the best possible number and state
of clusters for a database without requiring any information parameters [12]. A random point is provided to a cluster
which is then prepared incrementally without thinking about the cluster.
In DBCLASD, a cluster may be defined by three properties shown below:
1) Expected Distribution condition
2) Optimality Condition
3) Connectivity Condition
IV. METHODOLOGY
The below section of the paper describes the proposed work “Enhanced K-means Clustering using Euclidian
distance and Similarity Function”. Some meaningful and required results are being analyzed by the data mining tools
available. The facility of the data mining is being used in many applications in the real time for example customer
retention, education system, production control, healthcare, market basket analysis, manufacturing engineering,
scientific discovery and decision making etc.
In clustering the data objects available in the dataset is being grouped together such like that the data objects
with similar properties are together. In the case when the number of clusters in the dataset are less then the finer details
of the dataset might be missed for better representation. The clusters are being used for modeling the data. Data
modeling puts clustering in a historical perspective rooted in mathematics, statistics, and numerical analysis. As of
taking as per the machine learning the data mining is all about the hidden patterns, and also is unsupervised kind of
learning and sort of data concept is being shown by the representation using the data mining technique. When the
practical consideration of the data mining is being taken into account then there exist an unforgettable role of the data
mining tools in many applications like exploration of the scientific information, mining of text and information
retrieval, applications based on spatial datasets, analysis of internet, CRM, marketing, medical diagnostics,
computational biology, and many others.
In this work we have extended the Arpit Bansal et al. (2017), “Improved K-mean Clustering Algorithm for
Prediction Analysis using Classification Technique in Data Mining” The k-means methodology is being used for the
prediction of the data which are very much similar to each other. A function is being selected on the basis of the
relevancy of the function and along with the Euclidian distance is being used for clustering the data points. The
enhancement in the k-means methodology is being done on the basis of the normalization of the data. In the
enhancement two new features will be added. The first point is to calculate normal distance metrics on the basis of
normalization. The selection is being done on the basis of the majority voting which is considered as the second stage
in the data mining process.
Vol. 16, No. 5, May 2018
ISSN 1947-5500

The problem of accuracy and redundancy of the dissimilar points in the clusters remains in the improved k-
means for which new enhanced approach is been proposed which uses the similarity function for checking the
similarity level of the point before including it to the cluster.
A. Proposed Architecture
NO
YES
Start
Data Input
Compute Similarity
Check for
Similarity
Condition
Compute the Centroid of the cluster
Preprocessing of data
Compute Euclidean
Distance
Vol. 16, No. 5, May 2018
ISSN 1947-5500

Figure 1. Processing Architecture of the proposed technique.
Step by step execution of the proposed methodology (Enhanced K-means Algorithm):-
Step 1:- Data is being generated from user,
Step 2:- Preprocessing of data is being done like data cleaning for noise removal, data integration and data
visualization,
Step 3:- Enhanced k-means is applied over the data:
 Centroids of the clusters is being computed,
 Euclidian distance between centroid and other point is computed using the mathematical formula discussed
in step 1 of k-means algorithm,
 The point selected is being checked for similarity before including it to the cluster using the similarity
function,
𝑆𝑖𝑚𝑀𝑠𝑟(𝐶𝑖) = ∑ 𝑆𝑖𝑚𝐹𝑢𝑛𝑐(𝐶 𝐶 𝑖, 𝐶 𝑝 𝑗
)𝐶 𝐶 𝑖, 𝐶 𝑝 𝑗
∈𝐶 (1)
Where, 𝐶 𝐶 𝑖, is the centroid of the ith cluster,
𝐶 𝑝 𝑗
is the cluster point which is to be included,
𝐶 is the total clusters.
Step 4:- Check the data point for similarity if the similarity of the nearest point is high then go to step 5 else consider
the next closest point for similarity checking.
Step 5:- The point is being included in the clusters if it appropriately defines the cluster similarity.
B. Concept of Similarity
It is natural to ask what kind of standards we should use to determine the closeness, or how to measure the dissimilarity
or similarity between a pair of objects, an object and a cluster, or a pair of clusters. This segment of the paper talks
about the concept of similarity and also about the hierarchical clustering and the proximity between the clusters is
Close
Include the data
point into Cluster
Vol. 16, No. 5, May 2018
ISSN 1947-5500

being estimated. The single cluster is being represented by the prototype so that the cluster can be processed further
just similar to any other object present in the dataset.
A data object is described by a set of features, usually represented as a multidimensional vector. The features
may be quantitative or qualitative, continuous or binary, nominal or ordinal, which estimates the methodology. A
distance or dissimilarity function on a data set is represented to fulfill the following conditions.
Likewise, a similarity function is defined to satisfy the conditions in the following.
1- Symmetry:- S(xi, xj)=S(xj, xi)
2- Positivity:- 0<=S(xi, xj)<=1 for all xi and xj
3- S(xi, xj)=1 if xi=xj and it is termed as similarity matric.
For a data set with input N patterns, we can define an N*N symmetric matrix, called proximity matrix, whose
(i, j)th
element represents the similarity or dissimilarity measure for the ith
and jth
patterns (i, j =1,…….,N).
Typically, the function for distance estimation which is being utilized for estimating the features which are
just continuous, and also the similarity estimation is considered as the qualitative variables. The process of method
selection is just issue dependent. For binary features, a similarity measure is commonly used (dissimilarity measures
can be estimated by Dij=1-Sij). Suppose we use two binary subscripts to count features in two objects. n00 and n11
represent the number of simultaneous absence or presence of features in two objects, n01 and n10 and count the features
present only in one object. Then two types of commonly used similarity measures for data objects xi and xj are
illustrated in the following.
𝑆𝑖,𝑗 =
𝑛00+𝑛11
𝑛11+𝑛00+𝑤(𝑛10+𝑛01)
(2)
Where,
W=1, simple matching coefficient
W=2, Rogers and Tanimoto measure.
W=1/2, Gower and Legendre measure
The above equation is being used to estimate the similarity in any two data objects. The pairs which remains
un-matched are provided with some weights and which decides the similarity of the objects or data objects.
𝑆𝑖,𝑗 =
𝑛11
𝑛11+𝑤(𝑛10+𝑛01)
(3)
Vol. 16, No. 5, May 2018
ISSN 1947-5500

Where,
W=1, Jaccard coefficient
W=2, Sokal and Sneath measure.
W=1/2, Gower and Legendre measure
These measures focus on the co-occurrence features while ignoring the effect of co-absence. For nominal
features that have more than two states, a simple strategy needs to map them into new binary features, while a more
effective method utilizes the matching criterion
𝑆𝑖,𝑗 =
1
𝑑
∑ 𝑆𝑖,𝑗 (4)
where
𝑆𝑖,𝑗 = {
0 𝑖𝑓 𝑖 𝑎𝑛𝑑 𝑗 𝑑𝑜 𝑛𝑜𝑡 𝑚𝑎𝑡𝑐ℎ
1 𝑖𝑓 𝑖 𝑎𝑛𝑑 𝑗 𝑚𝑎𝑡𝑐ℎ
The proposed methodology starts from the point of getting data from the user as the dataset to be used is
dynamic in nature hence data is like changing very frequently. Data pre-processing stands for noise removal and other
related steps which can be used so as improve the quality of the data or representation of data. The centroid of the
clusters is then computed for each cluster’s as the number of clusters are predefined for the particular dataset.
For every centroid the Euclidean distance with the available data point or newly entered data point is being
computed and considered in the way of increasing the distance from the centroid. In the very next step the similarity
of the data point with the centroid is being evaluated for examining that how much the data point which is to be
included look like the centroid.
If the similarity of the data point selected on the basis of the shortest distance from the centroid has highest
similarity, then it is included in the cluster else the very next data point with respect to the distance is then selected
for similarity checking and hence the process keeps on searching till the time the most similar data point is found. As
the number of iterations will increase in the overall checking hence will impact the complexity of the overall process
but will provide better clusters in which data points included will be of similar properties.
V. ANALYSIS
Both real and the synthetic datasets are being experimented. The synthesized datasets are created with the help of data
generator, which follows the basic spirits of well-known IBM synthetic data generator for the generation of clusters.
The data size, the number of objects and the average size of the transactions are the important features in the obtained
data.
Vol. 16, No. 5, May 2018
ISSN 1947-5500

GUI screenshots
Vol. 16, No. 5, May 2018
ISSN 1947-5500

The results are represented in the form of clusters of the data points. As the number of objects are increased
the proposed methodology outperforms the related techniques for data mining of the large databases in the terms of
Vol. 16, No. 5, May 2018
ISSN 1947-5500

the efficiency of the clusters and also it is represented that the proposed work is best suited for mining the real dataset-
based applications.
Advantages of proposed methodology
The proposed methodology is advantageous with the other related techniques for mining the large databases in the
following aspects:
 The running time is stable for the complete range of number of clusters for the complete synthetic dataset.
 The methodology is more robust as compared to previous others as it have two analysis parameters (Euclidian
Distance and Similarity).
 There is a large gap in the execution time when the number of items is increased to an extent; hence out
performs other previous techniques for mining the large databases.
 As the execution time in the case of large number of items is efficient hence the proposed methodology is
quietly more suitable for the real data mining applications.
 The efficiency of the clusters is higher as the similarity is considered for the inclusion of the data points in
any cluster with respect to the evaluation done with the cluster centroid.
VI. CONCLUSION
The paper discus a novel technique for clustering of the dataset which enhances the k-means clustering which uses
similarity of the data objects with each other and also with the centroid of the cluster as well in each iteration. If the
similarity of the data objects selected on the basis of the separation from the centroid has highest similarity, then it is
included in the cluster else the very next data point with respect to the distance is then selected for similarity checking
and hence the process keeps on searching till the time the most similar data point is found. As the number of iterations
will increase in the overall checking hence will impact the complexity of the overall process but will provide better
clusters in which data points included will be of similar properties.
REFERENCES
1. Berkhin P. (2006), “Survey of clustering data mining techniques in grouping multidimensional data”, Springer, pp. 25–71.
2. R. Agrawal, J. Gehrke, D. Gunopulos, and P. Raghavan (1998), “Automatic subspace clustering of high dimensional data for data mining
applications”, in Proc. ACM SIGMOD Int. Conf. Management of Data, pp. 94–105.
3. Abbasi A, Younis M.(2007), “A survey on clustering algorithms for wireless sensor networks”, Computer Communications, vol. 30, pp.
14-15.
4. Arpit Bansal et al.(2017), “Improved K-mean Clustering Algorithm for Prediction Analysis using Classification Technique in Data
Mining”, International Journal of Computer Applications (0975 – 8887) Volume 157 – No 6, pp. 35-40.
5. T. Kanungo and D. M. Mount (2002), “An Efficient K-means Clustering Algorithm: Analysis and Implementation”, Pattern Analysis
and Machine Intelligence, IEEE Transactions on Pattern Analysis and Machine Intelligence. vol. 24, no. 7.
6. J. C. Bezdek (1981), “Pattern Recognition with Fuzzy Objective Function Algorithms”, New York: Plenum Press.
7. Zhang T, Ramakrishna R, Livny M.(1996), “BIRCH: An efficient data clustering method for very large databases”, Proceedings of the
ACM SIGMOD, International Conference on Management of Data, vol. 25(2), pp. 103–14.
Vol. 16, No. 5, May 2018
ISSN 1947-5500

8. Martin Ester, Hans-Peter Kriegel, Jörg Sander and Xiaowei Xu (1996), “A Density- Based Algorithm for Discovering Clusters in Large
Spatial Databases with Noise”, The Second International Conference on Knowledge Discovery and Data Mining (KDD-96), Portland,
Oregon, USA.
9. Levent Ertoz, Michael Steinback, Vipin Kumar (2003), “Finding Clusters of Different Sizes, Shapes, and Density in Noisy, High
Dimensional Data”, Second SIAM International Conference on Data Mining, San Francisco, CA, USA.
10. Nagpal, Pooja Batra, and Priyanka Ahlawat Mann (2011), “Comparative study of density-based clustering algorithms”, International
Journal of Computer Applications 27.11, pp. 421-435.
11. XU, X., ESTER, M., KRIEGEL, H.-P., and SANDER, J.1998, “A distribution-based clustering algorithm for mining in large spatial
databases”, In Proceedings of the 14th ICDE, Orlando, FL, pp. 324-331.
12. Berkhin P. (2006), “Survey of clustering data mining techniques in grouping multidimensional data”, Springer, 2006, pp. 25–71.
13. Tian Zhang, Raghu Ramakrishnan, Miron Livny (1996), “BIRCH: An Efficient Data Clustering Method for Very Large Databases”,
SIGMOD ’96 6/96 Montreal, Canada, pp. 103-114.
Vol. 16, No. 5, May 2018
ISSN 1947-5500

Improved K-mean Clustering Algorithm for Prediction Analysis using Classification Technique in Data Mining

Recommended

More Related Content

What's hot (16)

Similar to Improved K-mean Clustering Algorithm for Prediction Analysis using Classification Technique in Data Mining (20)

Recently uploaded (20)

Improved K-mean Clustering Algorithm for Prediction Analysis using Classification Technique in Data Mining