A Cluster Based Approach for Geographic Mapping

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395 -0056
Volume: 03 Issue: 02 | Feb-2016 www.irjet.net p-ISSN: 2395-0072
A CLUSTER BASED APPROACH FOR GEOGRAPHIC MAPPING
Piyali Sarkar
M.Tech C.S.E. Dept. of Computer Engineering
St.Vincent Pallotti College of Engg & Tech. Nagpur, India
Abstract- Today we can’t imagine our lives without cell phones
which have become part and parcel of our life. Technology today
had made life easier and better. Often a country’s technological
advancement is measured on the basis of how people can easily
communicate with each other. Now-a-days many application
software are being introduced in markets for various purposes to
make life easier. The apps which are widely used are the GPS
system or Google maps which are used for navigation purposes.
GPS systems can be used to calculate latitudes, longitudes and
altitudes of current location irrespective of their location. The
work which has been done is addition and only a small
contribution to the work done by researchers all around the
world in the field of geo-tagging. Here a dummy database is
created which includes landmarks, city and state and various
statistical processes are used to cluster data according to the
distance measured. All the three methods are compared and
analyzed on the basis of formation of fair clusters. It is
represented on map according to the database of longitudes and
latitudes and distance measures. Then the clusters are decided
whether the landmarks are present in the same city or not
according to a predefined value of radius. For instance, if the data
is in format India gate, Delhi; Lotus Temple, Delhi; Taj-Mahal,
Agra, Uttar-Pradesh; then India Gate and Lotus temple should be
present in the same city of Delhi. Taj mahal, present in Agra in
near to Delhi. But it should not be shown in the cluster. There
might be a possibility that within 100 m radius while
representing clusters it may be included. We have tried to solve
this problem and also tried to give accurate results as much as
possible. Also the data which is used is taken in sequential format
rather than random points.
Keywords—Geo-tagging, geo-coding, clustering, feature
extraction, statistical methods.
I. INTRODUCTION
Mobiles are no longer restricted to calls. They have been
used for various purposes like music, video, text,To-do lists and
many more. Moreover, an added advantage to all recent
technologies, they can be controlled and used by the end users.
We can state that mobile has become a tool which is widely used
to access the internet. Also they are used for playing games, for
television and GPS and also video calling and barcode reading. In
past few years technology has witnessed many researchers
proposing high number of navigation systems. Some of these
systems were built successfully and have been in use since for
outdoor applications namely GPS system or GLONASS (Global
Navigation Satellite System) which are widely used in transport
applications to find the shortest route to the destination and also
for pedestrian navigation.
Geo-tagging refers to the process of adding tags or
information, basically geographical information to any media
posted on social sites. Geo-tagging can be further classified as
Geo-Coding, Geo-Blogging and Geo-Microformat. Geo-Coding
refers to the process of taking non-coordinate based geographical
identifiers such as street address and searching associated
geographic coordinates. Geo-Blogging refers to the process of
tagging blog data to attract search engine users searching for a
particular blog which contains any geographical information.
Geo-microformat is a process where it represents a physical
location on the Earth using latitudes and longitudes. Here the
main focus is on Geo-coding and the mapping is done on the
principles of geo-microformat.
Geo-Coding or in other words forward geo-coding
refers to give meaningful description to a particular location such
as street address or postal address or place name by using
geographic co-ordinates such as latitudes, longitudes and
elevation for projection on map. Reverse Geo-coding refers to
give meaningful description to a geographic co-ordinates for
location or place name. Further Geo-coding can be classified as
interpolation and point-level geo-coding. The address
interpolation technique can be well described with the help of a
suitable example. Consider a street address 166 Pandey layout.
The entire locality ranges from 110-210. Let us consider that
even numbers fall on the west side and odd numbers fall on the
east side of the street. The work of geo-coding is to match an
address with a street block as every block contains a different
street range, and would interpolate or calculate the position. In
this example, 166 is almost half way from 110 towards 210. This
point would be on the west side of the street. A point will be
mapped at this location. But this process has no provisions to
solve the problem of ambiguity. The problem arises when the
street address would be 166 pandey layout and 166 E pandey
layout. However this problem of ambiguity can be solved by
defining proper address sets including city, state and country.
Another problem that interpolation faces is to map addresses
which have not been added to the database. Point level geo-
coding gives more accurate mapping by using centroid of a
particular building or land parcel and then maps the location.
This is more effective for flood determination.
Geocoding is the process of assigning a location, usually
in the form of coordinate values, to an address by comparing the
descriptive location elements in the address to those present in
the reference material.[1][2]
© 2016, IRJET | Impact Factor value: 4.45 | ISO 9001:2008 Certified Journal | Page 207

Geocoding, also called address matching, can also be
defined as the process of plotting geographic locations
(identifiers) for data that is stored in tabular format. One can use
geocoding to plot the geographic positions of a set of addresses in
a customer database or one can plot Latitude and Longitude
points recorded from a GPS unit. Thus, Geocoding requires a
database of properly formed address, a database of streets and a
set of rules for matching the addresses to the streets.
Addresses come in many forms, ranging from the common
address format of house number followed by the street name and
succeeding information to other location descriptions such as
postal zone or census tract. In essence, an address includes any
type of information that distinguishes a place.
When you type an address or a place name in the search box and
in return the map shows a marker at the place. The process of
associating an address or a place name with coordinates on the
map is called Geocoding. In a spatial database this is done as a
point layer with name of the place as an attribute to the point
location. This is one way of geocoding. For addresses, the
associated coordinates are not saved in a database directly, but
computed using a method called linear referencing. The start and
end addresses along a line segment are saved and intermediate
addresses are interpolated and the coordinates are calculated.
In some online mapping service, you may have seen
satellite imagery. When these images are captured from a
satellite or an airplane, they are just plain images, like
photographs. But to display these images on a map, they need to
be associated with map coordinates. This process is called Geo-
referencing. Once the image is associated with the map
coordinates it can be overlaid on top of street maps. For
georeferencing, you can use a GIS software such as ArcGIS or
QGIS to georeference an otherwise un-referenced image or
scanned maps, and load them into Oracle Spatial.[2]
The widely used technique is the GPS system. GPS
systems were first developed for military applications. But today
any handheld devices have GPS unit which can receive radio
signals that satellites transmit. Irrespective of our location atleast
four satellites are visible. GPS is used to provide location and
time information. GPS systems are accurate theoretically up to 14
nanoseconds but are not authenticated and also don’t provide
any provisions to avoid false data shown on maps. For example, if
someone feeds information about Parvati complex near Sadar,
inspite of Padole Square which is the actual address of the
mentioned place then it would show the same on the map.
Many companies have found geo-location or geo-
targeting technology to be of value in Internet advertising. Pay-
per-click search engines like Google and Yahoo offer the ability for
advertisers to deliver targeted advertising banners based on the
location of the website visitor's IP address. For marketers,
geocoding is critical in targeting specific demographics. Appending
demographic census track data to latitude and longitude
coordinates helps marketers target the right demographics - those
who would be most likely to respond to their offer or marketing
message. Insurance companies are relying increasingly on
geocoding techniques to help set premiums and make underwriting
decisions based on the physical location of the insured property.
Take Hurricane Katrina and storm-surge damage, for example.
Most insurance carriers have their own set of rules and criteria
when it comes to underwriting, such as property elevation and
determining the distance of the property from/to the coast. Such an
imprecise standard may leave carriers insuring properties that
may not be situated in a flood zone, but are actually in a storm-
surge zone - where the flood exclusions in their policies would not
apply. [3]
One of the most widespread uses of geocoding technology
is in store/dealer locators. Businesses use geo-coded data to
ascertain proximity to potential customers, distance to suppliers
and competitors, service areas and delivery routes. You've probably
experienced a locator lookup yourself - maybe to find a restaurant,
pet shop or the Sprint Nextel store nearest your home or
business. However, to adequately serve its 52 million customers,
Dominos often has multiple stores located within the same ZIP
Code. So which neighborhood store is closest to a particular
customer's home? Using a geocoding solution to power the Store
Locator on its website allows Dominos to turn the street addresses
of its stores into usable location information - so customers can
actually determine which store is closest in relation to their home
address. But what if the address is wrong? Without accurate
addresses, it would be difficult to obtain accurate geocoding. For
instance, a geocoding application might not recognize the
difference between 123 Pandey Layout and 123 S. Pandey Layout,
which could be two totally different addresses located miles apart.
A bad address diminishes the accuracy of a store locator - it's the
biggest reason why some store locators get it wrong. That's why
businesses are more proactive about integrating routines for
address verification with their geocoding initiatives.
II. LITERATURE SURVEY
The most widely used technique in the field of geo-tagging is
Chi square method. Chi square is used to define a relation
between two points if there is no relation between them. The
work of chi square can be defined as guessing. In context of
statistics, chi square is defined as the comparison or difference
between the observed and expected value. In context of text
mining chi square is used by considering a term ‘t’ in a document
and looking up to the probability of occurrence of ‘t’ in the given
document and probability that the term ‘t’ is present in other
document i.e. other than the preferred document. In geo-tagging

it is used as every city is counted and compared that how many
times it has occurred.[1][2]
The best way to describe this method let us take an example
of some cities from state Maharashtra and Madhya Pradesh. The
cities and their corresponding states are listed in the following
table 1:
City State
Nagpur Maharashtra
Bhopal Madhya Pradesh
Mumbai Maharashtra
Nasik Maharashtra
Aurangabad Maharashtra
Bilaspur Madhya Pradesh
Raipur Madhya Pradesh
Table 1. List of city and their respective states
The above table is a list of cities with their respective states.
As we observe, every city is a unique city. No city is repeated.
Therefore the observed value will be 1 for all cities as they occur
only once in a particular state and also state other than the one in
which they are present. Next is the expected value. First the
observed value is summed up. Here in this example it is 7. Now
there are two states shown in the above table. So the expected
value will be 7/9 for each cell. The expected value will be 0.7777.
This value will be same for all the cities as they have the same
observed value. This is shown in the table below :
City State Observed
Value
Expected
Value
Nagpur Maharashtra 1 0.7777
Bhopal Madhya Pradesh 1 0.7777
Mumbai Maharashtra 1 0.7777
Nasik Maharashtra 1 0.7777
Aurangabad Maharashtra 1 0.7777
Bilaspur Madhya Pradesh 1 0.7777
Raipur Madhya Pradesh 1 0.7777
Table 2. Expected and Observed Values
Now Chi Square can be calculated by the formula given
below:
𝟀2 =
Where ‘O’ is the observed value and ‘E’ is the expected value.
The difference is squared and divided by the expected value.
The pro of this method is that it is very easy to implement.
But the cons of this procedure are it lacks in accuracy. Accurate
clustering requires precise definition of closeness between a pair
of objects. Moreover chi square never works for continuous data
and is designed to analyze categorical data. Since the work of chi
square is guessing, accurate relations are not defined.
To overcome these drawbacks of chi square, KDE was
introduced. KDE estimates the density directly from the data
without assuming a particular form for the underlying
distribution.
Mathematically, a kernel is a positive function. This is
given by K(x; h) where h is the bandwidth parameter and this
function is controlled by ‘h’. [2] If the kernel form is ‘Gaussian’
the density estimate at a point ‘y’ within a group of points xi
where i=1,2,3,……,N is given by :
ΡK(y) =
The bandwidth act as a smoothing parameter
controlling the trade off between bias and variance in the result.
In context of geo-coding, KDE can be used with a valid
distance metrics though the results are only normalized by
Euclidean distance. For navigation purpose haversine distance is
used which gives much accurate circles on the sphere between
two points from their latitudes and longitudes. The haversine
distance can be calculated by the following equation:
Haversin (d/r) = haversine(φ2-φ1) + cos(φ1)cos(φ2)
haversine (λ2-λ1)
Where λ1 and λ2 are longitudes at point 1 and 2
respectively and φ1 and φ2 are latitudes at points 1 and 2
respectively. Further the haversine function is given as sin2 (θ/2).
Also, haversine (d/r)=h which is calculated as inverse sine
function given by r haversine-1(h) = 2r arcsine( where arcsine
is nothing but inverse sine. ‘r’ is the radius of the sphere. Here the
image given below shows a geo-spatial data representing two
different species in South American continent.

Figure 2. KDE with haversine distance
KDE lacks because it yields density estimates that have
discontinuities and weights equally all points regardless of their
distance to the estimation point. The basic principle of kernel
density estimation is suppose there is a point say ‘A’ and it is to
be estimated that whether it is possible to club within the cluster.
But the entire densities of the cluster or nearer points are
clubbed together. Even if point A should be within the cluster or
even if related then it is not included within the cluster as it is not
nearer to the entire cluster. Thus point wise consideration is not
considered. Also in KDE if the bandwidth is smaller it leads to
unsmooth density distribution. Performance degrades in higher
dimensions.
III. PROPOSED METHODOLOGY
Thus to overcome these drawbacks of Chi square and KDE cosine
similarity method has been used. In context of text mining cosine
similarity measure is used in the following manner: consider a
document such as
T= {The sky is blue
The sun is bright
The sun in the sky is bright.}
In cosine similarity, the first statement is compared with itself
and with other two statements. In the first statement when it is
compared with itself all the words are same i.e. the statement will
be same and its value will be 1. When the statement will be
compared with the second statement the value will be much less.
When it will be compared with the third statement there are
many similar words i.e. ‘The’ , ‘sky’, ‘is’ therefore the value will be
nearer to 1.
The documents will be ranked accordingly. If the
problem of synonymy occurs, cosine inspite of allowing partial
matching this problem can be overcome by using linear algebra
or word text or any geographical gazette/dictionary is used.
In context of geo-coding,
cosθ = / || || || ||
The cosine similarity calculates the cosine of angle
between two vectors. By this formula it can be stated that how
can two points be related by seeing the angle and not the
magnitude.
Cosine similarity is mostly used in high dimensions. For
instance consider a database which consists of cities and states.
Every city is a unique city within that particular state and other
than that state. Therefore a different dimension is assigned to
each city say in terms of latitudes and longitudes. Also, the cities
will be characterised by a vector where the value of each
dimension may depend on its occurrence in the entire database.
In this situation cosine gives how two points can be likely to each
other. Cosine similarity gains its popularity due to measurement
of cohesion in clusters in data mining. Also, cosine similarity is a
good support for positive spaces. But it does have a name in good
metrics. The reason behind this is that it doesn’t have triangle
inequality property. The triangle inequality property states that
any side of a triangle is less than the sum of other two sides of a
triangle. To get through this, angular similarity is used i.e.
converted to angular similarity.
IV. EXPERIMENTAL RESULTS
All the three methods are compared on the basis of accuracy and
elapsed time. Accuracy in terms of cluster formation is measured
for all the three statistical methods. The following table shows
the comparison on basis of elapsed time by all the three methods.
Methods Elapsed Time(Sec)
Chi Square 6.6196
Cosine Similarity 2.4329
KDE 4.9948
Table 3. Comparison on the basis of Elapsed time
The following graph is the pictorial representation of the above
table showing the most effective method in terms of elapsed time.
Figure 3. Graph representing the Elapsed time of all the
three methods

The values of first 16 records are as follows:
CITY
CHI
VALUES
COSINE
VALUES
KDE
VALUES
Achalpur 0.9208 0.0045 0
Achhnera 0.9208 0.0625 0.42
Adalaj 0.9208 0.0758 1.8
Adilabad 0.9208 0.0803 3.12
Adityapur 0.9208 0.0116 4.38
Adoni 0.9208 0.1205 6.3
Adoor 0.9208 0.1294 7.08
Adra 0.9208 0.1651 8.64
Adyar 0.9208 0.2007 9.54
Table 4. Calculated Values of all the three methods
There are total 1400 records of nearly all cities in different states.
Here the cities are unique cities and are considered in a matrix
formation while calculation of cosine similarity and KDE
representing number of rows as i and columns as j. The graph
shown below is for 50 records :
Figure 4. Graph showing the values for the methods
For clustering, a database with latitudes and longitudes will be
used. The points or the cities which will be plotted using latitudes
and longitudes will be shown on the map. They are clustered on
the basis of states right from the pre-processing step.
Accuracy is compared by comparing the values calculated by
each of the three methods out of the total. The analysis is as
follows :
Figure 5. Graph showing the accuracy of each method out of
the total cities considered
The clusters thus formed are shown in the figure below :
Figure 6. Clusters of all states
For a single city, the cluster is shown below :
Figure 7. Cluster of the city Kolkata which includes some of
its landmarks.
CONCLUSION
Cosine is known as one of the best metric as compared to Chi
square in terms of accuracy and time. It is more accurate because
Chi square guesses the relation but not cosine similarity. Some of
the problems which occur with cosine are also considered and
solved regarding synonymy and triangle inequality by using
appropriate measures. As compared to KDE, as it faces the
problem of considering individual points, cosine gives more
accurate results in such case. Both work in higher dimensions
where KDE performs more accurately but not with smaller

dimensions where cosine works far better. Also cosine has least
elapse time and also more accurate than other two methods.
REFERENCES
[1] S. Dinesh, K.S. Kannan, “Kernel Based Tagging Method Using
Spatial Paradigm,” International Journal of Computer Science and
Mobile Computing, IJCSMC, Vol.3, Issue. 11, November 2014, Pg. 73-
80.
[2] Olivier Van Laere, Jonathan Quinn, Steven Schockaert , Bark
Dhoedt, Member, IEEE, “Spatially Aware Term Selection for Geo-
tagging,” IEEE transactions on Knowledge and Data Engineering,
Vol. 26, No.1, January 2014.
[3]Y.Yang and J.O. Pedersen, “A Comparative Study On Feature
Selection in Text Categorization,” Proc. 14th Int’l Conf. Machine
Learning, pp. 412-420 , 1997.
[4] C. Hauff and G.J. Houben, “WISTUD at MediaEval 2011: Placing
Task,” Proc. Working Notes Of the MediaEval Workshop, 2011.
[5] Martha Larson, Mohammad Soleymani, Pavel
Serdyukov,“Automatic Tagging and Geotagging in video
collections and communities” ICMR,2011.
[6] Olivier Van Laere, Steven Schockaert, Vlad Tanasescu, Bart
Dhoedt, Christopher B. Jones, “ Geo-referencing Wikipedia
Documents Using Data From Social Media Sources,” ACM
Transactions on Information Systems
[7] S. Ahern, M. Naaman, R. Nair, and J. Yang, “ World explorer:
Visualizing aggregate data from unstructured text in geo-
referenced collections,” Proceedings of the SeventhACM/IEEE-CS
Joint Conference on Digital Libraries, May 2007

A Cluster Based Approach for Geographic Mapping

Recommended

More Related Content

Similar to A Cluster Based Approach for Geographic Mapping (20)

More from IRJET Journal (20)

Recently uploaded (20)

A Cluster Based Approach for Geographic Mapping