International Journal of Computer Trends and Technology (IJCTT) volume 10 number 2 Apr 2014

ISSN: 2231-2803 http://www.ijcttjournal.org Page102

Different Type of Feature Selection for Text Classification

M.Ramya
#1
, J.Alwin Pinakas
#2
(Institute Of Information Management,Kgisl / Bharathiar University,India)


ABSTRACT Text categorization is the task of
deciding whether a document belongs to a set of pre
specified classes of documents. Automatic
classification schemes can greatly facilitate the
process of categorization. Categorization of
documents is challenging, as the number of
discriminating words can be very large. Many
existing algorithms simply would not work with these
many numbers of features. For most text
categorization tasks, there are many irrelevant and
many relevant features. The main objective is to
propose a text classification based on the features
selection and pre-processing thereby reducing the
dimensionality of the Feature vector and increase the
classification accuracy. In the proposed method,
machine learning methods for text classification is
used to apply some text preprocessing methods in
different dataset, and then to extract feature vectors
for each new document by using various feature
weighting methods for enhancing the text
classification accuracy. Further training the
classifier by Naive Bayesian (NB) and K-nearest
neighbor (KNN) algorithms, the predication can be
made according to the category distribution among
this k nearest neighbors. Experimental results show
that the methods are favorable in terms of their
effectiveness and efficiency when compared with
other.
Keywords Feature selection, K-Nearest Neighbor,
Nave Bayesian, Text classification.
I. INTRODUCTION
Automated text classification is a
particularly challenging task in modern data analysis,
both from an empirical and from a theoretical
perspective. This problem is of central interest in
many internet applications, and consequently it has
received attention from researchers in such diverse
areas as information retrieval, machine learning, and
the theory of algorithms. Challenges associated with
automated text categorization come from many
fronts: one must choose an appropriate data structure
to represent the documents; one must choose an
appropriate objective function to optimize in order to
avoid over fitting and obtain good generalization and
one must deal with algorithmic issues arising as a
result of the high formal dimensionality of the data.

Feature selection, i.e., selecting a subset of
the features available for describing the data before
applying a learning algorithm, is a common
technique for addressing this last challenge. It has
been widely observed that feature selection can be a
powerful tool for simplifying or speeding up
computations, and when employed appropriately it
can lead to little loss in classification quality.
Nevertheless, general theoretical performance
guarantees are modest and it is often difficult to claim
more than a vague intuitive understanding of why a
particular feature selection algorithm performs well
when it does. Indeed, selecting an optimal set of
features is in general difficult, both theoretically and
empirically; hardness results are known, and in
practice greedy heuristics are often employed.
.







Fig 1: Text Classification Framework
1. Related Work
Several researchers have emphasized on the
issue of redundant attributes, as well as advantages of
feature selection for the Nave Bayesian Classifier,
not only for induction learning. Pazzani explores the
methods of joining two (or more) related attributes
into a new compound attribute where the attribute
dependencies are present. Another method, Boosting
Data
set
Pre
processing
Feature
Selection
Reduce the
Feature set
Apply
Classification
Performance
Measure
International Journal of Computer Trends and Technology (IJCTT) volume 10 number 2 Apr 2014
ISSN: 2231-2803 http://www.ijcttjournal.org Page103

on Nave Bayesian classifier has been experimented
by applying series of classifiers to the problem and
paying more attention to the examples misclassified
by its predecessor. However, it was shown that it fails
on average in a set of natural domain. Langley and
Sage use a wrapper approach for the subset selection
to only select relevant features for NB. Cardie uses
the attributes from decision trees in combination with
nearest neighbor methods. And in a domain for
discovering patterns in EEG-signals, Kubat,
Flotzinger, and Pfurtscheller tried the use of Decision
tree in feature selection for Nave Bayesian classifier.
And recently, Augmented Bayesian Classifiers was
introduced as another approach where Nave Bayes is
augmented by the addition of correlation arcs
between attributes.

II. PROPOSED METHODOLOGY
1. Preprocessing
The goal behind preprocessing is to
represent each document as a feature vector, that is,
to separate the text into individual words. In the
proposed classifiers, the text documents are modeled
as transactions. Choosing the keyword that is the
feature selection process, is the main preprocessing
step necessary for the indexing of documents.

This step is crucial in determining the
quality of the next stage, that is, the classification
stage. It is important to select the significant
keywords that carry the meaning, and discard the
words that do not contribute to distinguishing
between the documents.

2. Stop Word Removal
In most of the applications, it is practical to
remove words which appear too often (in every or
almost every document) and thus support no
information for the task. Good examples for this kind
of words are prepositions, articles and verbs like be
and go. If the box Apply stop word removal is
checked, all the words in the file swl.txt are
considered as stop words and will not be loaded. This
file contains currently the 100 most used words in the
English language which on average account for a half
of all reading in English. If the box Apply stop word
removal is unchecked, the stop word removal
algorithm will be disabled when the corpus is loaded.


3. Stemming Algorithm
Stemming is the process of grouping words
that share the same morphological root. E.g. game
and games are stemmed to the root game. The
suitability of stemming to Text Classification is
controversial. In some examinations, Stemming has
been reported to hurt accuracy. However, the recent
tendency is to apply it, since it reduces both the
dimensionality of the term space and the stochastic
dependence between terms.

3.1 Porter Stemming
Porter stemming algorithm is a process for
removing the commoner morphological ending words
in English [8]. Rules in porter stemming algorithm
are separated into five distinct steps:
1) Gets rid of plurals and -ed or -ing. eg-> caress
ponies -> ponities -> ti caress -> caress cats -> cat
2) Turns terminal y to i when there is another vowel
in the stem. eg happy->happi
3) Maps double suffices to single ones. so -ization ( =
-ize plus -ation) maps to -ize etc.
4) Deals with -ic-, -full, -ness etc. similar strategy to
step3.
5) Takes off -ant, -ence etc.

3.2 Lancaster Stemming
The new debugging options helped to solve the
mystery of why the original rules generated the stem
"abud" from "abusively":
<abusively> 100->abusive 13->abusiv 94->abuj 27-
>abud
Affix removal conflation techniques are
referred to as stemming algorithms and can be
implemented in a variety of different methods. All
remove suffices and/or prefixes in an attempt to
reduce a word to its stem.. The algorithms that are
discussed in the following sections, and those that
will be implemented in this project, are all suffix
removal stemmers.
III. FEATURE WEIGHTING
For many machine learning algorithms it is
necessary
to reduce the dimensionality of the feature space, if
the original dimensionality of the space is very high.
In most of the cases this improves not only the
International Journal of Computer Trends and Technology (IJCTT) volume 10 number 2 Apr 2014
ISSN: 2231-2803 http://www.ijcttjournal.org Page104

performance but also the accuracy of the
classification itself. Term Weighting is the process of
assigning values to all the terms in the corpus
according to their importance for the actual
classification part. Here, importance is defined as the
ability of the term to distinguish between different
categories in the corpus. Usually, the more important
a term is the higher is the assigned weight value.

1. Document Frequency (DF)
Simply measures in how many documents
the word appears. Since it can be computed without
class labels, it may be computed over the entire test
set as well. Selecting frequent words will improve
the chances that the features will be present in
future test
cases. It is defined as
1
( )
m
i
i
DF A
=
=

2. Mutual Information (MI)
The mutual information of two random
variables is a quantity that measures the mutual
dependence of the two random variables. MI
measures how much information the
presence/absence of a term contributes to making the
correct classification decision.
{ } { } 1,0 1,0
( , )
( , ) ( , ) ln
( ) ( )
k
k
k
f ck
k f c
f c
vf vc
P F v Ck V
MI F C p F v ck v
P F V P Ck V
e e
= =
= = =
= =


3. Information Gain (IG)
Here both class membership and the
presence/absence of a particular term are seen as
random variables, and one computes how much
information about the class membership is gained by
knowing the presence/absence statistics (as is used in
decision tree induction.
It is defined by following expression

It is frequently used as a term goodness criterion in
machine learning. It measures the number of bits
required for category prediction by knowing the
presence or the absence of a term in the document.

4. X2 Statistic (CHI)
Feature Selection by Chi - square testing is
based on Pearsons X 2 (chi square) tests. The Chi
square test of independence helps to find out the
variables X and Y are related to or independent of
each other. In feature selection, the Chi - square test
measures the independence of a feature and a
category. The null-hypothesis here is that the feature
and category are completely independent. It is
defined by,
2
, ,
2
(( ) ( , , ))
( , )
,
k Ck Ck F Ck
F Ck
F C F F
k
F Ck
N N N N N
x F C
N N N

=


5. Ngl Coefficient
The NGL coefficient is a variant of the Chi
square metric. It was originally named a `correlation
coefficient', name it `NGL coefficient' after the last
names of the inventors Ng, Goh, and Low. The NGL
coefficient looks only for evidence of positive class
membership, while the chi square metric also selects
evidence of negative class membership.
( , , , , )
( , )
F
F Ck F Ck F Ck
k
F Ck Ck
N N N N
NGL F C
N N N N

=

6. Term Frequency Document
Frequency
The tfidf weight is a method based on the term
frequency combined with the document frequency
threshold, it is defined as,
1 2 1 3 2 3 ( ) ( ( )) TFDF F n n c n n n n = + +
7. GSS Coefficient
The GSS coefficient was originally presented in
[GSS00] as a `simplified chi square function'. We
International Journal of Computer Trends and Technology (IJCTT) volume 10 number 2 Apr 2014
ISSN: 2231-2803 http://www.ijcttjournal.org Page105

follow [Seb02] and name it GSS after the names on
the inventors Galavotti, Sebastiani, and Simi.
( ) , , , , , F Ck F Ck F F Ck N N = Ck
k
Gss F c N N

IV. TEXT CATEGORIZATION
With the rapid growth of online information,
there is a growing need for tools that help in finding,
filtering and managing the high dimensional data.
Automated text categorization is a supervised
learning task, defined as assigning category labels to
new documents based on likelihood suggested by a
training set of labeled documents. Real-world
applications of text categorization often require a
system to deal with tens of thousands of categories
defined over a large Taxonomy. Since building these
text classifiers by hand is time consuming and costly,
automated text categorization has gained importance
over the years.

1. KNearest Neighbor Classifier Algorithm
K-Nearest Neighbor is one of the most
popular algorithms for text categorization. K-nearest
neighbor algorithm (k-NN) is a method for
classifying objects based on closest training examples
in the space. The working of KNN can be detailed as
follows first the test document has to be classified the
KNN algorithm searches the nearest neighbors
among the training documents that are pre classified.
The ranks for the K nearest neighbors based on the
similarity scores are calculate using some similarity
measure such as Euclidean distance measure etc., The
distance between two neighbors using Euclidean
distance can be found using the given formula the
categories of the test document can be predicted
using the ranked scores. The classification for the
input pattern is the class with the highest confidence;
the performance of each learning model is tracked
using the validation technique called cross validation.
The cross validation technique is used to validate the
pre determined metric like performance and
accuracy.


While using kNN algorithm, after k nearest
neighbors is found, several strategies could be taken
to predict the category of a test document based on
them. A fixed k value is usually used for all classes in
these methods, regardless of their different
distributions. Equation (1) and (2) below are two of
the used strategies of this kind method.


where d
i
is a test document, x
i
is one of the neighbors
in the training set,y(x
j
,c
k
) indicates whether
x
j
belongs to class c
k
,sim(d
i
,x
j
) and is the similarity
function for d
i
and x
j
. Equation (1) means that the
predication will be the class that has the largest
number of members in the k nearest neighbors;
whereas equation (2) means the class with maximal
sum of similarity will be the winner.
2. Naive Bayesian Classifier Algorithm
The Naive Bayes classifiers are known as a
simple Bayesian classification algorithm. It has been
proven very effective for text categorization.
Regarding the text categorization problem, a
document d 2 D corresponds to a data instance,
where D denotes the training document set. The
document d can be represented as a bag of words.
Each word w 2 d comes from a set W of all feature
words. Each document d is associated with a class
label c 2 C, where C denotes the class label set. The
Naive Bayes classifiers estimate the conditional
probability P(c|d) which represents the probability
that a document d belongs to a class c. Using the
Bayes rule, we have


The key assumption of Naive Bayes classifiers is that
the words in the documents are conditionally
independent given the class value, so that


A popular way to estimate P(w|c) is through
Laplacian smoothing:


2
1
( , ) ( )
D
i
Dist X Y Xi Yi
=
=

International Journal of Computer Trends and Technology (IJCTT) volume 10 number 2 Apr 2014
ISSN: 2231-2803 http://www.ijcttjournal.org Page106

where n(w, c) is the number of the word positions
that are occupied by w in all training examples whose
class value is c. n(c) is the number of word positions
whose class value is c. Finally, |W| is the total
number of distinct words in the training set.

V. PERFORMANCE METRIC
The evaluation of a classifier is done using
the precision and recall measures .To derive a robust
measure of the effectiveness of the classifier It is able
to calculate the breakeven point, the 11-point
precision and average precision . to evaluate the
classification for a threshold ranging from 0 (recall =
1) up to a value where the precision value equals 1
and the recall value equals 0, incrementing the
threshold with a given threshold step size. The
breakeven point is the point where recall meets
precision and the eleven point precision is the
averaged value for the precision at the points where
recall equals the eleven values 0.0, 0.1, 0.2... 0.9, 1.0.
Average precision refines the eleven point
precision, as it approximates the areabelow the
precision/recall curve.
The 11-point average precision is another measure
for representing performance with a single value. For
every category the
i
CSV threshold is repeatedly
tuned such that allow the recall to take the values
. At every point the
precision is calculated and at the end the average over
these eleven values is returned [Sebastiani02]. The
retrieval system must support ranking policy.
VI. EXPERIMENTAL RESULTS
Data Set 1: Self Made
For the development used a small self-made corpus
that contains standard categories such as Science,
Business, Sports, Health, Education,
Travel, and Movies. It contains around 150
documents with the above mentioned categories.
Data Set 2: The Reuters 21578 corpus
The second corpus included for the development is
Reuters 21578 corpus. The corpus is freely available
on the internet (Lewis 1997). Uses an XML parser, it
was necessary to convert the 22 SGML documents to
XML, using the freely available tool SX (Clark
2001). After the conversion I deleted some single
characters which were rejected by the
validating XML parser as they had decimal
values below 30. This does not affect the
results since the characters would have been
considered as whitespaces anyway.
Table 1: Performances of two classification
algorithms
Da
ta
Set
Knn SVM
Mic
ro-
Avg.
Macro-Avg. Mic
ro-
Avg.
Macro-Avg.

Pre
=Re
c= F
1
Pr
e
R
ec
F
1
Pre
=Re
c= F
1
Pr
e
R
ec
F
1
Re
ute
rs
71.6
9

67
.1
0

66
.5
7

66
.0
2

72.3
5

70
.7
4

77
.2
4

77
.9
6

20
ne
ws
fro
up
70.1
2

67
.8
8

67
.5
6

66
.2
9

75.2
4

75
.1
7

78
.4
9

78
.1
1

Avg.

70.9
0
67
.4
9
66
.4
3
66
.1
5
73.7
9
72
.9
5
77
.8
6
78
.0
3

To evaluate the effectiveness of category
assignments by classifiers to documents, the standard
precision, recall, and F 1 measure are used here.
Precision is defined to be the ratio of correct
assignments by the system divided by the total
number of the systems assignments. Recall is the
ratio of correct assignments by the system divided by
the total number of correct assignments.
These scores can be computed for the binary
decisions on each individual category first and then
be averaged over categories. Or, they can be
computed globally over all the n*m binary decisions
where n is the number of total test documents, and m
is the number of categories in consideration. The
former way is called macro-averaging and the latter
International Journal of Computer Trends and Technology (IJCTT) volume 10 number 2 Apr 2014
ISSN: 2231-2803 http://www.ijcttjournal.org Page107

micro-averaging. It is understood that the micro-
averaged scores (recall, precision, and F 1) tend to be
dominated by the classifiers performance on
common categories, and that the macro-averaged
scores are more influenced by the performance on
rare categories.
VII. CONCLUSION
Analyzed the text classification using the
Naive Bayesian and K-Nearest Neighbor
classification. The methods are favorable in terms of
their effectiveness and efficiency when compared
with other classifier such as SVM. The advantage of
the proposed approach is classification algorithm
learns importance of attributes and utilizes them in
the similarity measure. In future the classification
model can be build that analyzes terms on the
sentence, document.
REFERENCES
[1] A Fuzzy Self-Constructing Feature Clustering Algorithm for
Text Classification Jung-Yi Jiang, Ren-Jia Liou, and Shie-Jue
Lee, Member, IEEE TRANS ON Knowledge and Data Eng.,Vol
23,No.3,March 2011

[2] J. Yan, B. Zhang, N. Liu, S. Yan, Q. Cheng, W. Fan, Q. Yang,
W. Xi,and Z. Chen, Effective and Efficient Dimensionality
Reduction for Large-Scale and Streaming Data Preprocessing,
IEEE Trans.Knowledge and Data Eng., vol. 18, no. 3, pp. 320-333,
Mar. 2006.

[3] H. Li, T. Jiang, and K. Zang, Efficient and Robust Feature
Extraction by Maximum Margin Criterion, T. Sebastian,
S.Lawrence, and S. Bernhard eds. Advances in Neural Information
Processing System, pp. 97-104, Springer, 2004.
[4] D.D. Lewis, Feature Selection and Feature Extraction for Text
Categorization, Proc. Workshop Speech and Natural
Language,pp. 212-217, 1992.

[5]http://kdd.ics.uci.edu/databases/
reuters21578/reuters21578.html. 2010.

[6] H. Kim, P. Howland, and H. Park, Dimension Reduction in
Text Classification with Support Vector Machines, J. Machine
Learning Research, vol. 6, pp. 37-53, 2005.

[7] F. Sebastiani, Machine Learning in Automated Text
Categorization, ACM Computing Surveys, vol. 34, no. 1, pp. 1-
47, 2002.
[8] H. Park, M. Jeon, and J. Rosen, Lower Dimensional
Representation of Text Data Based on Centroids and Least
Squares, BIT Numerical Math, vol. 43, pp. 427-448, 2003.