Two Layer k-means based Consensus Clustering for Rural Health Information System

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
© 2017, IRJET | Impact Factor value: 5.181 | ISO 9001:2008 Certified Journal | Page 45
Two Layer k-means based Consensus Clustering for Rural Health
Information System
Ms. Archana Singh1, Prof.Dr.V.H.Patil2
1PG scholar, Department of computer Engineering, Matoshri College of Engineering and Research Centre,
Maharashtra, India
2Professor, Department of computer Engineering, Matoshri College of Engineering and Research Centre,
Maharashtra, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - This paper presents a data clustering approach
using two layer k means based consensus clustering. This
algorithm helps to partitionstheheterogeneous dataandform
the sub clusters for main clusters to find efficient decision
making in rural health information system. In this paper we
provide a systematic study of two layer k means based
consensus clustering. It helps in both completeandincomplete
datasets. Experimental results on MCTS dataset demonstrate
that two layer k-means based consensus clustering is highly
efficient and comparable to the rural health information
system. This algorithm shows high robustness to incomplete
basic partitioning with many missing values.
Key Words: Consensus clustering, k-mean. Sub cluster,
classification
1. INTRODUCTION
In Traditional K-means algorithm, a datum X will be
assigned to the cluster C where the distance between X and
the cluster center of C is minimal, comparing to the distances
between X and the cluster centers of otherclusters.However,
the abnormal data may be assigned to most of clusters but
normal data are classified into a few clusters.
Consensus clustering,alsoknownasclusterensembleor
clustering aggregation, aims to find a single partitioning of
data from multiple existing basic partitioning[3].Ithasbeen
widely recognized that consensus clustering can help to
generate robust clustering results,handlenoise,outliersand
sample variations, and integrate solutions from multiple
distributed sources of data or attributes [4].
The main objective to developthissystemthathelps,the
rural officers to store and upload the data directly to the
cloud server which will help them in reducing the paper
work. According to the existing scenario the rural officersgo
and collect the information of the rural infants and pregnant
lady to providethem with proper vaccination.Thedetailsare
collected and after few days the details are stored into the
server and the further process of vaccination is carried out,
this process take time to get the output. The vaccination is
provided to the pregnant lady till the birth of baby with a
regular period of time. The required medicine is also
provided to the babies till the age of 5 years.
This project is provides proper messages to the families
about the necessary of taking the vaccination atpropertime,
this application will provide the notifications to the families
about the date, time and location of the vaccination facility
provided by the government. This will help in getting the
proper count of infants and pregnant ladies in a villagethose
who have taken the vaccination which will help the rural
officer to know who has taken and number of counts
remaining to take the vaccination from a particular village.
It will notify the families about the funds that are
released from the government to the pregnant ladies and to
new born babies in rural area for the pregnancy and for
further medical support. This help in improving the
nutritional and health status of children’s in theagegroup 0-
5 years to enhance the capability of the mother to look after
the normal health and nutritional needs of the child through
the proper nutrition and through health education.
To develop this system we are going to use two layer k-
means based Consensus clustering to take decision on the
scattered or heterogeneous data. Before work on this
algorithm first we know about basic concepts of Consensus
Clustering and two layer K-means.
2. REVIEW OF LITERATURE
This chapter presents the study of the existing systems
and proposed work related to the proposed System. The
purpose of the literature survey is to identify information
relevant to project work andthepotential andknownimpact
of it within the project area.
Rural Health Information System is an existing system
which works on a mission of providing the information, to
the rural areas about the vaccination required for infants
and pregnant women. The Rural Health Information System
has rather had adverse impact on health system, all services
are paralyzed and health supervisors are sitting in front of
office computers and making data entries. Currently Rural
Health Information System portal is an absolute online
version and for poor internet connectivity it cannot be
operated properly. Adding a feature for offline data entry
process followed by online uploading to the serverwould be
useful. As no IEC activity can be conducted duringdata entry
Volume: 04 Issue: 07 | July -2017 www.irjet.net p-ISSN: 2395-0072

period, which may lead to some primary health issues. Till
the data entry work outsourcing is done, they arepromoting
record maintenance at overall primary health care services.
Regular feedback system will be established in this system.
This system will be in a position to benefit the entire health
care system incorporating more useful indicators.Hence, we
summarized that the existing system is inefficient, time
consuming, poorly managed, and lacking flexibility. This
solutions very useful as the solution is inherently
distributive.
3. SYSTEM OVERVIEW
3.1 Consensus Clustering
Consensus clustering, also known as cluster ensemble
or clustering aggregation, aimstofinda single partitioningof
data from multiple existing basic partitioning[3].Ithasbeen
widely recognized that consensus clustering can help to
generate robust clustering results,handlenoise,outliersand
sample variations, and integrate solutions from multiple
distributed sources of data or attributes [4]. Here we briefly
introduce the basic concepts of consensus clustering and
formulate the research problem of this paper.
Let X = {x1, x2, ··· ,xn} denote a set of data
objects/points/instances. A partitioning of X into K crisp
clusters is represented as a collection of K subsets of objects
in C = {Ck|k = 1,··· ,K}, with CkTCk0 = ∅, ∀k 6= k0, and SK k=1
Ck = X, or as a label vectorπ = hLπ(x1),··· ,Lπ(xn)i, where
Lπ(xi) maps xi to one of the K labels in {1,2,··· ,K}. We also
use some conventional math notations as follows. For
instance, R, R+, R++, Rd, and Rnd denote the sets of reals,
non-negative reals, positive reals, d-dimensional real
vectors, and n×d real matrix, respectively. Z denotes the set
of integers, and Z+, Z++, Zd and Znd are defined analogously.
For a d-dimensional real vector x, kxkp denotes the Lp norm
of x, i.e., kxkp = p qPd i=1 xp i , |x| denotes the cardinality of
x, i.e., |x| =Pd i=1 xi, and xT denotes the transposition of x.
The gradient of a single variable function f is denoted as ∇f,
and the logarithm of based 2 is denoted aslog.Ingeneral,the
existing consensus clustering methods can be categorized
into two classes, i.e., the methods with or without global
objective functions [9]. In this paper, we are concerned with
the former methods, which are typically formulated as a
combinatorial optimizationproblemasfollows.Givenrbasic
partitioning of X (a basic partitioning is a crisp partitioning
of X by some clustering algorithm) in Π = {π1,π2,··· ,πr}, the
goal is to find a consensus partitioning π such that
(1)
is maximized, where Γ : Zn++ × Znr++ 7→ R is a consensus
function, U : Zn++×Zn++ 7→ R is a utility function, and wi ∈
[0,1] is a user-specified weight for πi, with Pr i=1 wi = 1.
Sometimes a distance function, e.g., the well-known Mirkin
distance [5], rather than a utility function is used in the
consensus function. In that case, we can simply turn the
maximization problem into a minimizationproblemwithout
changing the nature of the problem. Consensus clusteringas
a combinatorial optimization problem is often solved by
some heuristics and/or some Meta heuristics.Therefore, the
choice of the utility function in Eq. (1) is crucial for the
success of a consensus clustering, since itlargelydetermines
the heuristics to employ. In the literature, some external
measures originally proposed for cluster validity have been
adopted as utility functions for consensusclustering,e.g.,the
Normalized Mutual Information [3], Quadratic Mutual
Information [6], and Rand Index [8]. These utility functions
of different math properties pose computational challenges
to consensus clustering.
3.2 Two Layer K-mean Algorithm
In the traditional K-means algorithm, a datum X will be
assigned to the cluster C where the distance between X and
the cluster center of C is minimal, comparingtothedistances
between X and the cluster centers of other clusters.
However, if there are outliers or noisy data in a data set, the
abnormal data may be assigned to most of clusters but
normal data are classified into a few clusters. In Fig. 1, the
red dots are the abnormal data which are only a few data
relative to normal data and are classified to two clusters but
the vast majority of normal data are classified to only one
cluster. It is usually non-helpful for future analysis. In data
clustering, when the data in one cluster are quite different,
the respective features of the cluster cannot precisely
represent all the data in this cluster. According to this
requirement, a two-layer K-means algorithm is proposed to
improve traditional K-means algorithm. The two-layer K-
means algorithm contains three steps:
A. Data normalization,
B. Cluster center initialization, and
C. Two-layer clustering.
A. Data normalization
In distance-based classification, a small variationin
one feature is probably more influencing than a big
variation in other feature when computing the distance
of two data. It is necessary to normalize every feature
value of each feature dimension to a specific range. This
stage is to transform all variables in thedata toa specific
range. Let S={X1, X2, …, XN} be a data set consisting of N
data, Xi be the i-th data in S, and (xi1, xi2, …, xid) be the
features of Xi. For each feature value xid is normalized
by (2).

(2)
B. Initial cluster center
In this step, a most discrepant initial clustercentermethodis
proposed to determine the initial cluster centers for K-
means algorithm. It uses the biggest discrepant data as the
initial cluster centers. Let the distance dik oftwodata C1and
C2:
(3)
where wj and rj are the given constants. The
algorithm first decides two data C1 and C2, where
C1=Xi and C2 = Xk, and
(4)
After that, it computes the data: C3 which is
farthest from C1 and C2, C4 which is farthest from C1,
C2 and C3, CK which is farthest from C1, C2, …, and
CK-1, where C1, C2, …, and CK are in S and are
considered to the initial cluster centers of the K
clusters.
C. Two-Layer Clustering
K-means algorithm uses a cluster center to representthe
data of the cluster. If the dissimilarity of data is big in a
cluster, the cluster center cannot describe all of the data in
the cluster. For example, there are two clustersC1wherethe
dissimilarity of data is very big and C2 where the
dissimilarity of data is very small. Then enter a data point x
which is closer to the border of C1 but far from C2. On the
contrary, x is far from the cluster center of C1 but closer to
C2.
As Fig. 2(a), x will be mistakenly classified to the cluster
C2 in the traditional K-means algorithm. However, whenthe
data in C1 are separated into several smaller clusters (sub
clusters), x will be pointed to one of sub-clusters in the
cluster C1 as Fig. 2(b) so x will definitely belong to cluster
C1. This study proposes two-layer K-means algorithm to
improve the traditional K-means algorithm due to theabove
problems. Each cluster is subdivided into several sub-
clusters in two-layer K-means algorithm and then combines
with the traditional K-means algorithm for data clustering.
Firstly, two-layer K-means algorithm adopts the data
normalization step and the initial cluster center step and
then cluster data set S into K clusters using traditional K-
means algorithm. Next assumingthatCGistheG-thclusterof
data set S, and two-layer K-means algorithm usestraditional
K-means algorithm to divided the CG into KG sub-clusters
with CG1, CG2, …, G GKC . To input a data point (v1, v2… vd),
the data point is detected belongs among certain sub-
clusters of CG , then it will be attributed to an element of CG.
Next let CGgj is the j-th dimension value of g-th sub-cluster
center in the G-th cluster, the distance dis between the data
point and each sub-cluster center as follows:
(5)
Figure 2. Two Clusters and a data point-x
Figure 2. Two Clusters and a data point-x
Fig.1 Cluster Example
Fig.1 Cluster Example

The sub-cluster is not only closest to the data point,
but also belongs to the cluster CG, and then the data
point is classified into CG.
4. SYSTEM ANALYSIS
In proposed system we will consider two one istwo
layer k-means and second is consensusclustering.It will
help to make decision on heterogeneous data which
were not considered in existing system.Inthisapproach
heterogeneous data is considered as input and are used
to test and validate the results.
As Shown in figure 3, the heterogeneous data is
taken as a input to system. This data iscollectedthrough
Asha's daily collection of information from different
regions. After that Consensus clustering is applied on
this data to make basic clusters. On these clusters two
layer k-mean algorithm is applied for making clustersof
sub clusters. These are nothing but the more nearest to
neighbor solution which are used to gain the efficient
results.
Fig. 4 explains that the flow of the MCTS using two
layer k-means based Consensus clustering to make
effective and correct decision on heterogeneous data
to send notifications or textual messages to pregnant
women for their better health.
In this our system there are three basic parts or
module:
1) Main admin
2) MCTS Officer
3) User Side (Baby’s care and Mom’s Care
In first part i.e. Main admin has web panel to handle
the following tasks, admin has rights to allocate MCTS
officer in particular part of country and track the work
of MCTS officers and the data inserted by the officers.
The admin has control to send proper notification to
user.
In second module, the MCTS officers also called
ASHA have to work on field. They are having application
to register pregnant women or new born baby. This will
reduces the paper work of MCTS officer which they are
doing currently.
Third module is important i.e. User Application
which will be used by end users. These users include
pregnant women and new born baby mothers.Pregnant
women will get proper notification for dosage andthere
timing during pregnancy period. New baby born
mothers will get notifications for vaccinations.
In this system we are going to apply the twolayerk-
means based consensus clusteringalgorithmonthedata
gathered by MCTS officers. The notification will fired on
the decision of algorithm applied on that data.
5. MATHEMATICAL MODELLING
Let S be a technique for two layer k means based
consensus clustering;the equationproposedwill befrom the
fundamental principles ofk meansandconsensusclustering.
Figure 4. Functional model of proposed
System
Figure 3. Proposed System Architecture

Input:
Cj represents the set of clusters
CKj represents the set of Sub Clusters
Where,
Cj = c1,c2,c3...cn
CKj = ck1,ck2,ck3...ckn
Output:
: The consensus partitioning
Process:
S={ D, X(b), K, pi}
Where,
S= System.
D= Healthcare Dataset
Where Cj, CKj are arbitrary dataset which is actual
input for system
X (b) = clusters
K=Two layer k means based consensus
Approach
Km = k-means clustering
2Km= 2 layer k means
pi= The consensus partitioning
6. RESULT ANALYSIS
After comparing the classifier performance against
all three data mining model, quiet interesting results
were discovered as shown below. We used a test bed
consisting of a number of real world data sets obtained
from UCI repositories like Iris, Ecoli and Wine.
Table 1 shows the complete summary of the
performance comparisons as executiontimeofthethree
data mining models used for this research work on
different datasets. Also Table 2 shows the complete
summary of the performance comparisons as accuracy
of the three data mining models used for this research
work on different datasets.
Table 1: Comparison of Execution Time (in ms)
We do the comparison of Two layer K-mean consensus
clustering algorithm with K-mean and Two layer k-mean in
terms of execution efficiency. Table 1 shows the runtime
comparison of the three methods where we observe that
proposed algorithm two layer k-mean based consensus
clustering algorithm is fastest among three methods.
Also table 2 demonstrated that accuracy of proposed
algorithm is higher than the other algorithms in different
datasets.
Table 2: Comparison of Accuracy (in percentage)
Algorithm Iris Ecoli Wine MCTS
K means 96.67 30.01 25.6 76.25
2 layer k
means
96.68 30.02 25.7 79.24
2layer k-
mean
Consensus
98.9 31.06 28.9 85.51
Algorithm Iris Ecoli Wine MCTS
K means 2 2 1 16
2 layer k
means
4 4 2 14
2layer k-
mean
Consensus
1 1 1 14

We established the general theoretical framework
of two layer k-means based consensus clustering and
provided the corresponding algorithm. Experiments on
real world datasets have demonstratedthattwolayerk-
mean based consensusclusteringhashigh efficiencyand
shows the robust performances.
7. CONCLUSION
In this paper, two layer k-means consensus
algorithm is proposed to improvethetwolayerk-means
and traditional k-means algorithm. It can give a better
accuracy rate of data clustering than other k-means
algorithm. The two layer k-means consensus algorithm
is used to strengthen the accuracy of data clustering.
Two layer k-means based Consensus Clustering for
Rural health information system used for classifyingthe
groups and sending the vaccination notification in the
form of textual messages and reminders to the families
of infants and pregnant women, according to the
vaccination dates periodically by using their registered
identification number.
8. ACKNOWLEDGEMENT
I want to thank all people who help me in different
way. Especially I am thankful to my guide and HOD
Prof.Dr.V.H.Patil for her continuous support and
guidance in my work.
REFERENCES
[1] Junjie Wu, Member, IEEE, Hongfu Liu, Hui Xiong, Senior
Member, IEEE, Jie Cao, Jian Chen, Fellow,IEEE,"K-means
based Consensus Clustering: A Unified View",IEEE
Transaction On Knowledge And Data Engineering, vol.
xx, no. xx, December 2013.
[2] Chen-Chung Liu1,Shao-Wei Chu2, Shyr-Shen
Yu4,*,Yung-Kuan Chan3,"A modified K-means
Algorithm- Two Layer K-means Algorithm",2014 Tenth
International Conference on Intelligent Information
Hiding and Multimedia Signal Processing.
[3] A. Strehl and J. Ghosh, “Cluster ensembles — a
knowledge reuse framework for combining partitions,”
JMLR, vol. 3, pp. 583–617, 2002.
[4] N. Nguyen and R. Caruana, “Consensus clusterings,” in
ICDM, 2007.
[5] B. Mirkin, “The problems of approximation in spaces of
relationship and qualitative data analysis,” Information
and Remote Control, vol. 35, p. 1424–1431, 1974
[6] A. Topchy, A. Jain, and W. Punch, “Combining multiple
weak clusterings,” in ICDM, 2003, pp. 331–338.
[7] Z. Lu, Y. Peng, and J. Xiao, “From comparing clusterings
to combining clusterings,” in AAAI, 2008, pp. 361–370.
[8] J. Wu, H. Xiong, and J. Chen, “Adapting the right
measures for k-means clustering,” in KDD,Paris,France,
2009, pp. 877–886.
[9] J. Wu, H. Xiong, C. Liu, and J. Chen, “A generalization of
distance functions for fuzzy c-means clustering with
centroids of arithmetic means,” TFS, vol. 20, no. 3, pp.
557–571, 2012.
[10] P.-N. Tan, M. Steinbach, and V. Kumar, Introduction to
Data Mining. Addison-Wesley, 2005.

Two Layer k-means based Consensus Clustering for Rural Health Information System

More Related Content

What's hot (18)

Similar to Two Layer k-means based Consensus Clustering for Rural Health Information System (20)

More from IRJET Journal (20)

Recently uploaded (20)

Two Layer k-means based Consensus Clustering for Rural Health Information System