A Deep Analysis on Prevailing Spam Mail Filteration Machine Learning Approaches

International Journal of Trend in Scientific Research and Development (IJTSRD)
Volume 4 Issue 6, September-October 2020 Available Online: www.ijtsrd.com e-ISSN: 2456 – 6470
@ IJTSRD | Unique Paper ID – IJTSRD33261 | Volume – 4 | Issue – 6 | September-October 2020 Page 1160
A Deep Analysis on Prevailing Spam Mail
Filteration Machine Learning Approaches
Anu1, Ms. Preeti2
1M Tech Student, 2Assistant Professor,
1,2Meri College, Sampla, Haryana, India
ABSTRACT
In this work, we have reviewed the issue of spam mail which is a big problem
in the area of Internet. The growing size of uncalled masse-mail (orspam)has
produced the requirement of a dependable anti-spam filter. Now a days the
Machine learning (ML) proedures are being employed to spontaneously filter
the spam e-mail in an effective manner. In this work, we have reviewed some
of the prevalent ML approaches (such as Rough sets, Bayesian classification,
SVMs, k-NN, ANNs and Artificial immune system) and of their use fullness in
the issue of spam Email taxonomy. We have provided the depictions of the
procedures and the divergence of their enactment on the basis of the quantity
of Spam Assassin.
KEYWORDS: Spam e-mail, Machine learning Procedures and Filtering
How to cite this paper: Anu | Ms. Preeti
"A Deep Analysis on Prevailing Spam Mail
FilterationMachineLearningApproaches"
Published in
International Journal
of Trend in Scientific
Research and
Development(ijtsrd),
ISSN: 2456-6470,
Volume-4 | Issue-6,
October 2020,
pp.1160-1167, URL:
www.ijtsrd.com/papers/ijtsrd33261.pdf
Copyright © 2020 by author(s) and
International Journal ofTrendinScientific
Research and Development Journal. This
is an Open Access article distributed
under the terms of
the Creative
CommonsAttribution
License (CC BY 4.0)
(http://creativecommons.org/licenses/by/4.0)
1. INTRODUCTION
With the increase in poularity of the Internet, we have also
faced new types of threats such as virus attack and
unwanted profit-making mass quantity of e-mails popularly
known as spam. These types of e-mails consume time,
communication bandwidth and storage space. In spite of so
many firewalls, this issue has been growing day by day.
According to some prevalanet survey, less than 50 % of all
emails are spam which cost the internet users a huge price
per year. In this scenario, an automated e-mail filteration
mechanism can be the most suitable technique for defying
spam at any time. Therefore there is a tuff fight among
spammers and anti-spammers. In previous times, the spam
was filtered through detction and blockage ofaddresswhich
send spam e-mails or through filtaration of messageshaving
specific keywords or advertisements. In order to mislead
filteration techniques, the spammers have stratedtoemploy
numerous tricks to disable the filtering techniques such as
use of different sender IP addresses or to affix / suffix
haphazard symbols to the beginning /end of the subject line
of the message [1].
The most popular methods of spam mail filteration are
knowledge engineering (KE) and machine learning (ML). In
KE based methods, we specifie a collection of rules to
identify spam emails as well as normal e-mails. This
collection of rules can be specified by the user, or by the
software vendorwhichbestowsa specificrulecenterdspam-
filtering software. In order to be successful, these rules
should be continuously renewed and preserved, which is
always not possible or it can be inconvenient for many
persons. After many successful Machine learning (ML)
approach, it was concluded that Machine learning methods
are better than knowledge engineering methods;asthesedo
not need to specify any instructions [2]. Rather, in machine
learning methods, there are a set of training samples which
are a group of pre specified e-mail posts. Next, an explicit
algorithm has been utilized to acquiretheclassificationrules
from these e-mail posts. In recent times, various Machine
learning methods (e.g. artificial immune system, J48
classifier, support vector machines (SVM), Naïve Bayes and
Neural Networks) have been extensively reviewed and can
be used for e-mail filtering. This work emphasized on the
analysis fn filtering of spam mail and protecting normal e-
mails.
This paper has been planned as follows: section 1 contains
introduction of the paper, section 2 synopsizetheassociated
work considering several ML procedures, section3provides
a common theoretic explanation on the ML approaches and
section 4 present conclusion of the paper.
2. LITERATURE SURVEY
A filtering method is needed for the categorization of e-mail
into ham or spam. The authors [3] have proposed a spam
email filteration method through distinctive features
selection process to categorize the emails into normal and
spam. After the pre-processing of the dataset (English and
Malay email) descriptions, they have used TF-IDF andrough
IJTSRD33261

International Journal of Trend in Scientific Research and Development (IJTSRD) @ www.ijtsrd.com eISSN: 2456-6470
set theoritical technique. Next, they have applied machine
learning approach for the categorization and obtain good
performance.
Authors inn [4] have proposed a new ML based method for
the categorization of email data [4]. The implementation of
the algorithmic contains KNN and Naïve Bayes algorithms
and presenting applicable outcomes in case of application of
algorithms on pre-processed dataset.
Additionally, authors in [5] have planned an ontology
centered email filtering approach.Theuseddatasethasbeen
categorized via J48 decision tree centered method. In
pursuance of testing the outcomes received after the
categorization, a RDF linguistic centered ontology was
produced by Jena.
Authors in [6] have used FFN network focused technique to
recognize the spam emailsandtoprecisetheoutcomes.They
have used Krill Herd algorithm to train the dataset which is
uniformly allocated into two splits for the training and
testing reasons.
AUTHOR FINDINGS/ METHODOLOGY RESEARCH GAP
Guzella T, Caminhas
W. M.- 2000 [7]
In this paper, the authors have projected an
inclusive analysis of latest growthsintheusesof
ML processes for Spam filtering. They have
emphasized on both textual as well as image
centered methods.Ratherthantakeintoaccount
Spam filtering as a customary classification
issue, they have highlighted the significance of
reflecting particular features of the problem,
specially perception meaning for the designing
of different filters.
They have analyzed the problem faced in revising
a classifier bestowing to the bag-of-words
illustration and a key transformation among
ancient naive Bayes models. They have concluded
the significant development in this field and
pointed out different characteristics whichareyet
to be discovered, particularly in additional
reasonable estimation situations.
Levent Özgür,
Tunga Güngör, and
Fikret Gürgen –
2004 [8]
The authors have proposed vigorous anti-spam
cleaning approaches for differentvernacularsin
common and specifically for Turkish, founded
on ANN and Bayesian classifiers. The
procedures are flexible and have twin modules.
The major module works on the morphology
and the next module categorizes the mails
through the roots. They have considered single
layer and multi layer perceptron (MLP) (ANN
tructures), and the inputs to the networks are
controlled by binary and probabilily
proototypes. They have used three approaches
binary, probabilistic, and advance probabilistic
for Bayesian classification prototypes.
The anti-spam filtering mechanism which has
been defined in this work consists of two
components: MM and LM. They have designed a
turkish morphological invetigation method. A
particular word in Turkish may relate to a idiom
built up of numerous words so Turkish is known
as agglutinative language. Due to this, the
morphological examination of turkish language is
more difficult than the investigation in other
languages such as Hindi.
In pursuance of determining the origin wordsthat
can be served as the aspects of the classification
procedure, they have used the ideat of shared
information. The aspect vectorhasbeendescribed
as a set of acute words that can be employed in
categorization. Initially, the candidate words (the
words of the training data) has been recognised.
Enrico Blanzieri,
Anton Bryl -2007,
[9]
In this work, the authors have evaluated an
example focused on spam filter supported by
the SVM-NN classifier theory. They have
combined the concepts of SVM and kNN.
Initially, to mark a message the classifier
discovers k nearest marked messages, and next
a SVM prototype has been trained for these k
samples and employed to mark the unidentified
sample. The authors have compared SVM-NN
with SVM and k-NN and presented the results.
To find the adjacent neighbors and the linear
kernel for SVM, they have usedEuclideanmetricin
their experiments. All messages were deliberated
as an unordered list of strings disjointed by gaps.
Occurrence of a particular token in a specific
section of a message, specifically in the region of
the header, was reflected as a binary aspectofthat
message. Next, the utmost recurrent characters in
the training dataset were choosed and utilized.
Therefore, every message has been characterized
through a vector of the binary descriptions.
The authors have assessed a learning centered
spam filter. The proposed method has overtake
SVM considerably over the minute dimensions of
the feature domain.
Mawuena Glymin
and Wojciech
Ziarko- 2007 [10]
This work exhibits a fundamental summary of
methods used for spam recognition through
probability assessment table focused analytical
data modelling. The main emphasis is on the
presentationof the solution which joins modest
methods with particular heuristics to make
comprehensive rough estimates of spam and
genuine e-mails via the VPRSM rough set
technique. Experimentations were organizedto
The spam recognition system working has been
allocated into two stages: a training stage and a
categorization stage. The rough set centered
machine learning has been used in the training
stage and it is used for the pool of pre-categorized
e-mails.
In the categorization stage, hierarchy of trained
decision table was employed which forecast the
choice class of receiving e-mails.

investigate the uses of VPRSM for creating a
smart worker for spam filteration.
They have conducted investigationofpossibility
of utilizing the hierarchy of VPSRM probability
tables for spam recognition filter.
For training purpose, they have collected e-mails
from hotmail platform and tested the
determination prototype.
Hsu Wei-Chih,Tsan-
Ying Yu- 2009 [11]
The SVM is a dominantcategorizationmethodof
data mining and has been effectively used in
several physical world applications.
The classification performanceiseffected bythe
selection of parameters of SVM specifically
through training process. Though, selection
parameter in SVMhasbeennormallyrecognized
by knowledge or simple grid search. They have
suggested Taguchi technique for improvement
of grid search and employed toenhancetheSVM
centered Spam Filteration prootootype.
It is simple to realize the orthogonal arrays
without repetition. A physical world mail
dataset was chosed to show the efficiency and
possibility of the process. The experimentation
exhibits that the Taguchi technique may
discover the operative prototype with high
categorization precision and good sturdiness.
The authors have discovered the finest blend of
two coefficients (C, γ) of SVM using Taguchi
technique. While creating spam filteration
prototype, thecoefficientsofSVMwereconsidered
as influence factors. After the selection of the
coefficients of SVM, they verified the results of
classification and equated with grid search.
The SVM is a vigorous supervised learning model
which is founded on the organized hazard
lowering standard from arithmetical learning
concept. SVM has amazing enactment for text
classification process.
Ibrahim Eldesoky-
2009 [12]
Artificial Immune System (AIS) prototype has
been stimulated from the normal immune
system. The authors have proposed the AIS
prototype which fulfill the spam filteration
procedure. The contents of the e-mail messages
has been employed for both training as well as
testing procedures.
The words which make the e-mail are weighted
and employed in estimating the empathy amid
the antibody and an antigen. To remunerate the
cells that properly identifythespame-mail,they
have initiated a learning phase. Rather than
using clonal assortment, negative assortment
was employed for the training period and it
caused improved testing enactment owing to
the condensed quantity of sensors. As equated
with other methods, the fraction of training
group elements is very less than others and
hence it confirms the improved enactment. The
procedure was verified against the global
Masses of spam and normal mails.
The detector or sensor is the key module of the
spam immune scheme and these are known as
antibodies (AB). The ABs are required to confirm
the subjects of an e-mail posts.
To transform an email messageintoanAB,initially
they have computed the native weight of every
word of the email. The native weights were
computed in a standardized way through totaling
of the existences of every word in relation to the
number words in the email. The anticipated
affinity assessment method is supportedby native
weights rather than totals of the analogous words
of the two emails.
Next, they have checked this affinity against a
standard value to indicate the type of the email.
Though, the artificial prototype has been trained
for both spam and desired elements, but it discard
the cells which erroneously identify new antigens
through the testing progression.
Loredana F.,
Camelia L.,Rodica P.
-2010 [13]
This work represents a fresh spam recognition
filter, whichconglomeratesnumerousattributes
which are prominently based on frequency of
words, and usesof the kNNprocesstocategorize
the emails.
The recommended architecture has been
separated into two key modules: one for the
collection of the data and for the extraction of
the aspects of all messages and another for the
Classification of the emails and interpretationof
the results.
Initially, an email has benn selected to assemble
the data for analysis. They have designed spam
traps to attract spammers.
For feature extraction, they have anayzed the
message on the basis of different features such
as length of message, quantity of responses and
frequency of words. An email and these mined
features signify the sample of the data set. The
authors havemergedthetokenizationtechnique
for the contents of the message with a curbing
The authors have classified the messages with the
kNN method which is supported on a group of
characteristics mined from the properties and
contents of the email. The trained set has been
resampled to the utmost suitable size and certain
class dissemination has been decided by
numerous experimentations. The proposed
mechanism achieves a continuous renewal of the
dataset and the record of maximum commonly
words that exist in the email. Moreover, they have
provided a feedback choice concerning
nonclassified messages and suggested to be
accomplished at a fixed rate conditional to the
availability of the resources.

procedure.
Tiago A. Almeida ·
Jurandy Almeida ·
Akebo Yamakami -
2011[14]
Here, the authors have considered the
enactment of distinct term-selection methods
using various separate prototypes of Naive
Bayes (NB) spam filters. They have designed
the experiments which were meticulously
designed to confirm statistically precise results.
Furthermore, they have performed an
investigation which concern
the magnitudes normally used to assess the
excellence of spam filters. Lastly, they have also
investigated the advatages of working the MCC
as a standard of enactment.
Initially the authors have showed an enactment
assessment of numerous term based approaches
in decline of dimensionality for the spam filtering
field by classifiers according to the Bayesian
decision scheme. They have accomplished the
assessment
of the enactment attained by dissimilar Naïve
Bayes spam filters used to categorize messages
from six famous, actual, public and huge e-mail
datasets.
They have also offered the MCC as the estimation
measurement factor which offeres an extra stable
estimation of the forecast than TCR, particularly if
the two groups are of dissimilar dimensions.
Yuanchun Zhu and
Ying Tan. - 2011
[15]
Stimulated from the living immune system,the
authors have recommended a local
concentration (LC) inspired feature mining
method for spam filteration. This method is
capable of extracting the location realted
information through messages by converting
every region of a message to an analogous LC
aspect. They have implmented two schemes of
the LC method by a static length and a varying
length sliding windows. A generic LC model has
been designed to include the LC methodintothe
complete practice of spam filteration.Twokinds
of detector groups have been producedthrough
term selection approaches and a clear penchant
threshold. Afterward a sliding window hasbeen
accepted to allocate the message into separate
regions. Once segmentation of the Message is
complete, next they have calculatedthestrength
of detectors and measure the aspect for every
native region. Conclusively, theyhavecombined
all the aspects of local areas as a aspect vectorof
the message.
To assess the anticipated LC systeml, they have
performed numerous experimentations on 5
standards masses through the cross confirmation
process. It has been proven that the LC method
works well with 3 term selection techniques.
When compared with the universal concentration
centerd method and the existing bag-of-words
technique, the LC technique has better enactment
in context of both precision and measurement.
Moreover, they have verified thattheLCtechnique
is vigorous for messages of different lengths.
Noemí Pérez-Díaz,
David Ruano-Ordás,
José R. Méndez,Juan
F. G., Florentino F. –
2012 [16]
The authors have reviewed and combined the
preceding methods and new substitutes for
smearing the rough set (RS) mechanism over
the spam filtering space through describing the
three distinct rule implementation techniques
such as MFD), LNO and LTS. Keepinginmindthe
aim of properly evaluatingthecorrectnessofthe
anticipated procedures, theypreciselysolveand
review important queries for suitableprototype
approval such as corpus assortment, pre-
handling and representative problems, and
distinctive exact standard processes.
The experimentation conducted through
numerous implementation methods for
choosing suitabledecisionrulesproducedbyRS.
Their anticipated procedures outdid other
popular anti-spam filteration procedures like
SVM, Adaboost and distinct sorts of Bayes
classifiers.
This work provides a complete revision about the
uses of RS as key classifier for business spam
filteration. They have presented and investigated
different schemes via RSs along with a realistic
analysis about their applications, drawbacks and
advantages.
After analysis, they have concluded that most of
the preceding works give an idea only using
corpora and an insufficient pre-handling. Keeping
this in mind, the authors have carried out a
different investigation for a huge, fresh and non-
managed corpus which was supplied by the team
of Spam Assassin.
As approved by the results received, the RS-
centered methods are continually an appropriate
substitute of Naïve Bayes classifier, SVM and
Adaboost method.
The proposed MFD heuristic reaches the finest
precision when matched with LNO and LTS.
Yazdan Jamshidi-
2016 [17]
For spam filtering,thiswork presentstheusages
of Interval’s Number KNN (INKNN). Later,this
method was defined as a lattice data set
expansion of KNN method.
A populace of spam e-mails was displayed using
an IN. Then proposed classifier was utilized to
differntitate spam e-mails from hams.
To examine the efficiency of INKNN, they have
performed wide experimentation over public
This work bestows a NN classification procedure
for spam filteration founded on probability
explained interval number and lattice concept.
A real word benefit of lattice concept is the
capability of modelling both indeterminate info
anddissimilar kinds of lattice-ordered data.
The suggested method is able to deal with
different kind of data. It is able to manage both
locations and intermissions. Thereisa fastprocess

SpamAssassin corpus. The outcomes presents
that the INKNN is capable of achieving the
advanced enactment as comapared to other
previous advanced ML methods.
of learning in the suggested method and hence it
can be used in different areas where the data is so
enormous that the examination process becomes
time devouring.
The key benefit of using interaval number is the
lodging of grainy information. The experiment
outcomes validate the efficiencyofouranticipated
prototype.
Ali Shafigh Aski,
Navid Khalilzadeh
Sourati– 2016 [18]
This work defines 3 ML procedures to separate
spam from hams with less error rates and
extreme competence through a multilayer
perceptron prototype. Numerous extensively
employed methods are decision tree (DT)
classifier, multilayer perceptron (MP) and NB
classifier. These methods are useful for training
datasets either in the type of spam or hams. At
last, they have discussedthe resultsofmeasured
methods are inspected in context of theplanned
prototype.
The proposed method is based on guidelines for
appropriate recording in context of the
competence of instructions. The instructionswere
delivered in three types: a) informationanalysisof
email header, b) tallying of keywords, and c) key
content of the message.
Priti S., Uma B. –
2018 [19]
This work is meant to suggest a ML based fused
bagging method by applying the two ML
procedures: NB and J48 (DT)) for the spam
email recognition. Here, the dataset has been
shared into dissimilar groups and provided as
input to all algorithms. They have conducted
various experiments and the obtained results
were compared in context of accuracy, memory,
exactness, f-measurement, true and false
negative/positive rate. One experiment was
conducted using Naïve Bayes and another using
J48 methods. One more experiment was
conducted through hybrid bagged method. The
complete precision of 87.5% attained by the
hybrid bagged method of spam mail detection
system.
On the basis of the idea of entropy, the J48 method
is a multcast decision tree classifier. It forms
decision trees of the training dataset. The DT is
created through J48 and relies on thetrainingdata
attribute values for the categorization of the fresh
data value. It adhere to the idea of dividing the
data into numerous groups,everyaspect qualityof
data can be utilized to make a decision. The
technique works iteratively till every data
attribute has been managed and classified.
Abdul J. S., Asif K.,
Bharanidharan S.,
Sami A., KrishnanK.,
Mirjam J. and Friso
D. – 2019 [20]
Spam mails are also popular as non-self/
unwanted moneymaking or malevolent mails.
These mails are composed and forwarded to
disturb either a person or a company ora unitof
peoples.
Other than promotion, the mails may have
associations with phishing or malware hosting
websites and can sneak private info.
In this work, the authors have studied the
efficacy of employing a NSA process for
inconsistency recognition smeared to spam
filteration. The proposed method has an
extreme enactment and a low false recognition
rate.
The proposed model is trained by forming a
memory of the previous conduct ofspamemails.It
forbids the similar type of conduct for future
incoming messages, because the model has been
trained against a certain conduct. This procedure
is known as Negative Selection (NS).
The strategy of the NSA has been founded on the
self and nonself discernment conduct of the
mammalian learned immune system.
The aim is to develop a prototype of irregularities,
variations, or unaware data throughproductionof
patterns which do not relate to or equal to a
prevailing body of accessible patterns. The NSA
determines gap among regularity and irregularity
by having information of selfandnon-selfconduct.
TABLE: 01 LETRATURE REVIEW / GAP
3. CLASSIFICATION OF MACHINE LEARNING IN E-MAIL
The classification task of E-mails is generally allocated into numerous sub-tasks. Initially the collection and demonstration of
Data are most problemsatic, next, feature selection and reduction of e-mail try to lessen the size for the residual stages of the
task.
Naïve Bayes classifier technique: Its works on the dependent incidence and the likelihood of an incidence happening in the
upcoming time which may be recognised from the preceding occuring of the analogous incidence. This method may be
employed to categorize spam e-mails. Here the likelihoods of words perform the key rule. Ifanywordstakeplacefrequentlyin
spam but not in normal e-mail, then this inward e-mail is perhaps a spam. Figure 1 shows the working of classifier method.

Fig 1: An instance of Naïve Bayes classifier
K-NN classifier technique: It is measured as an instancecenteredclassifier.Herecomparisonisdoneovertraining documents
instead of an specific class illustration, like the class profiles are employed by other classifiers. It means actually there is no
training stage. For the classification of a fresh document, the k most identical documents (neighbours) are located and if a
sufficient big share of them have been allocated to a definite class, the fesh document has also been allocated to this class.
Moreover, using customary indexing approaches cause the discovery process of the nearest neighboursveryfast.Todtermine
the class of the message (spam or ham), message which are close to the class are searched quickly. Figure 2 shows the kNN
based system classification [13].
Fig 2: KNN based spam mail filteration
Artificial Neural Networks (ANN) classifier technique:
It is also recognized as Neural Network (NN) and it is a computation prototype inspired by biological NN. It includes an
interrelated group of artificial neurons. An ANN isa flexiblesystemwhichmodifiesitsconfigurationaspertheinformationflow
within the artificial network throughout a learning stage. The ANN is founded on the law of learning by instances. Moreover,
there are two standard types of the NNs, perceptron and the MP. Figure 3 shows the working of ANN for spam mail
filteration[20].
Fig 3: working of ANN for spam mail detection
SVM classifier technique: It is founded on the idea ofdeterminationplanesthatdescribeitslimitations.Adeterminationplane
is splits among a group of entities having dissimilar groups participations. The SVM patterning procedure discovers an
optimum hyper plane with the highest gap to split two groups. It needs resolving the optimization problem. WorkingofSVMis
shown in figure 4 [21].

Fig 4: SVM classification of e-mail.
Artificial Immune System (AIS) classifier technique:
The normal immune system protects the person against dangerous illnesses and viruses. It is able to recognize and eliminate
any different cell or particle. To accomplish its job, the immune organism has grown over complicated pattern identification
and reply procedures follow several differential routes.
The fundamental task of natural immune organism is to differentiate among self and injurious non-self components.
The pathogens are the injurious non-self components of specific concern. Consequently theAIScandistinguishbetweenselfof
legitimate email and a non-self of illegitimate email. Its working is shown in figure 5 [12].
Fig 5: working of Artificial Immune System based spam filteration
Rough sets (RS) classifier technique:
these sets are capable of computing the diminutions of
information schemes which mayhavecertaincharacteristics
that are unrelated to the goal idea (means: determination
attribute), and certain surplus attributes. Diminution is
required to produce modest valuable in formations from
itself. It is a least subset of provision attributes in relevance
to determination attributes. The RS structure works as
follows: With the the received e-mails, initialy, we should
choose the maximum suitable attributes to employ for
classification. Then the input dataset is altered into a
decision structure and splitted into the training and the

testing datasets. A classifier will be persuaded from the
training dataset and smeared to the testing dataset to attain
enactment assessment.
4. CONCLUSION
In this work, we have reviewed certain well konwn machine
learning techniques and their suitability to the issueofspam
e-mail categorization. We have presented the research gaps
between the machine learning approaches towards spam e-
mail filteration. We have presented brief descriptions of the
popular machine learningmethodslikeNaïveBayesclassifie,
KNN, ANN, rough sets, and artificial immune systems. In
context of precision, we conclude that the Naïve bayes and
rough sets approaches are very satisfyingin enactmentamid
the other approaches. There is a need of extra research to
increase the enactment of the Naïve bayes and Artificial
immune system either by amalgam system or by resolution
of the feature dependence problem.
5. REFERENCES
[1] L. Cormack, Gordon. Smucker, Mark. Clarke, Charles
"Efficient and effective spam filtering andre-ranking
for large web datasets" Information Retrieval,
Springer Netherlands. January 2011, pp. 1-24.
[2] Guzella, T. S. and Caminhas, W. M.”A review of
machine learning approaches to Spam filtering.”
Expert Syst. Appl., 2009.
[3] M. Mohamad, and A. Selamat, "An evaluation on the
efficiency of hybrid feature selection in spam email
classification", In: Proc. of 2015 International
Conference on Computer, Communications, and
Control Technology (I4CT), Kuching, Sarawak,
Malaysia, pp.227-231, 2015.
[4] A. Harisinghaney, A. Dixit, S. Gupta, and A. Arora,
"Text and image basedspam email classificationusing
KNN, Naïve Bayes and Reverse DBSCAN algorithm",
In: Proc. of 2014 International Conference on
Optimization, Reliabilty, and InformationTechnology
(ICROIT), Faridabad, Haryana, pp.153-155, India,
2014.
[5] S. Youn, and D. McLeod, "Efficientspamemail filtering
using adaptive ontology." In: Proc. of Fourth
International Conference on InformationTechnology,
Las Vegas, NV, USA, pp.249-254, 2007.
[6] H. Faris, and I. Aljarah, "Optimizing feed forward
neural networks using Krill Herd algorithmfore-mail
spam detection", In: Proc. of IEEE Jordan Conference
on Applied Electrical Engineering and Computing
Technologies (AEECT), Amman, Jordan, pp.1-5, 2015
[7] Guzella T, Caminhas W, (2000) A review of machine
learning approaches to spam filtering. Exp Syst Appl,
vol. 36, issue 7, pp.10206–10222, 2000.
[8] Levent Özgür, Tunga Güngör, and Fikret Gürgen,
“Spam Mail Detection Using Artificial Neural Network
and Bayesian Filter”, in International Conference on
IntelligentData EngineeringandAutomatedLearning,
pp. 505-510, 2004.
[9] Enrico Blanzieri and Anton Bryl, “Instance-Based
Spam Filtering Using SVM nearest Neighbor
Classifier”, American Association for Artificial
Intelligence, www.aaai.org, pp. 441-442, 2007.
[10] Mawuena Glymin and Wojciech Ziarko, “Rough Set
Approach to Spam Filter Learning”, in RSEISP '07:
Proceedings of the international conferenceonRough
Sets and Intelligent Systems Paradigms, pp.350–359,
june 2007.
[11] Hsu Wei-Chih and Tsan-Ying Yu, “E-mail Spam
Filtering Using Support Vector Machines with
Selection of Kernel Function Parameters”, in Fourth
International Conference on Innovative Computing,
Information and Control, 765-767, 2009.
[12] M. H. Haggag and I. E. Fattoh, “Artificial Immune
System for Spam Filtering”,IJICIS,Vol.9,No.2,pp.117-
129, JULY 2009.
[13] Loredana Firte, Camelia LemnaruandRodica Potolea,
“Spam Detection Filter using KNN Algorithm and
Resampling”, in International Conference on
Intelligent Computer Communication and Processing
(ICCP), 2010 IEEE, pp. 27-33.
[14] Tiago A. Almeida, Jurandy Almeida and Akebo
Yamakami, “Spam filtering: how the dimensionality
reduction affects the accuracy of Naive Bayes
classifiers”, J Internet Serv Appl,pp. 183-200, 2011.
[15] Yuanchun Zhu and Ying Tan, “A Local-Concentration-
Based Feature Extraction Approach for Spam
Filtering”, IEEE Transactions on Information
Forensics and Security, vol. 6, no. 2, pp.1-25,2011.
[16] Noemí P., David R., José R. M., Juan F.G.andFlorentino
F., “Rough sets for spam filtering: Selecting
appropriate decision rules for boundary e-mail
classification”, Applied Soft Computing, pp. 3671–
3682, 2012.
[17] Yazdan Jamshidi, “A nearest neighbour classifier
based on probabilistically/possibilistically intervals’
number for spam filtering”, Int. J. Soft Computing and
Networking, Vol. 1, No. 1, pp. 4-16, 2016.
[18] Ali Shafigh Aski and Navid Khalilzadeh Sourati,
“Proposed efficient algorithm to filter spam using
machine learning techniques”, Pacific ScienceReview
A: Natural Science and Engineering, pp. 145-149,
2016.
[19] Priti Sharma and Uma Bhardwaj, “Machine Learning
based Spam E-Mail Detection”, International Journal
of Intelligent Engineering & Systems, Vol.11, No.3, 1-
10, 2018.
[20] Jyh-Jian Sheu, Yin-Kai Chen, Ko-Tsung Chu, Jih-Hsin
Tang and Wei-Pang Yang, “An intelligent three-phase
spam filtering method based on decision tree data
mining”, Security and communication networks, vol.
9, issue 17, 401304026, 2016.
[21] Shradhanjali and Toran Verma, “E-Mail Spam
Detection and Classification Using SVM and Feature
Extraction”, international Journal of Advance
Research, Ideas and Innovations in Technology, vol 3,
Issue 3, pp. 1491-1495.

A Deep Analysis on Prevailing Spam Mail Filteration Machine Learning Approaches

More Related Content

What's hot (20)

Similar to A Deep Analysis on Prevailing Spam Mail Filteration Machine Learning Approaches (20)

More from ijtsrd (20)

Recently uploaded (20)

A Deep Analysis on Prevailing Spam Mail Filteration Machine Learning Approaches