View of Detection of Unwanted Messages and Fraudulent User Identification on Social Network

(1)

1041 http://annalsofrscb.ro

Detection of Unwanted Messages and Fraudulent User Identification on Social Network

Dr. R. SenthamilSelvi¹,Dr. S. Mohana², Prof. S. Uma Maheswari³, Dr. R. Pushpalakshmi⁴, Prof.L.Amudha⁵

1Department of CSE,Saranathan College of Engineering,Trichy, [email protected]

2Department of CSE, Saranathan College of Engineering, Trichy, [email protected]

3Department of CSE,K.Ramakrishnan College of Technology,Trichy, [email protected]

4Department of IT, PSNA College of Engineering& Technology, Dindigul, [email protected]

5Department of CSE, K.RamakrishnanCollege of Engineering,Trichy, [email protected]

ABSTRACT

Nowadays, social networking sites play a vital role for users to communicate and sharing information with other users online. Statistical analysis show that, on average, people spend most of their time on Facebook and Twitter which are constantly maintaining their position in most-viewed social networking sites than other sites.

Usually many social networking users share their personal information and interest to other users publically.

This leads to violation of privacy and attracts many intruders to steal their personal information and use them in an unauthorized manner. So there emerges a major concern in securing their personal details from fake ids and cyber security has become very crucial. In this paper, spammer identification technique on social networks like Facebook,Twitter are addressed. It has been implemented based on URL used by the user, trending topics and fake content shared by the users and fake users. Various kinds of features which are posted by the users are identified from which the nature of the user is detected as spam or not. Statistical studies proved that, this spammer detection schemes improves the detection accuracy rate constantly

1.Introduction

It is very simple to acquire any data from any source over the world viaWeb. Expanded social destinations interest grants clients togetherrich measure of data and information about clients.Colossalvolumes of information accessible on these destinations likewise draw the consideration of phony clients[1]. Twitterhas quickly become an online hotspot for procuring on going data about clients. Twitter isan Online Social Network (OSN). In this,clients share everything without exception, for example, news, opinions, and even the indispositions^[27].A few contention scan be held over various points,forexample, governmental issues, current under takings,and significant occasions.Tweets by clients are passed to her/his supporters immediately.

It permits them for extending got data at highly expensive level. The necessity of examining and contemplating clients practices in online social stages are intensified with OSNs development.

Individuals without huge data based OSNs may be deceived by fraudsters without much stretch.

There is additionally an interest to battle what's more, place a control on individuals who use OSNs just for commercials and therefore spam others' records.It shows the way to breach the privacy of individual users and be a magnet for many intruders to take their private information and use them in an unauthorized way^[28].

[32][33]Recently, the discovery of spam in long rangeinformal communication destinations pulledin the consideration of specialists. Spam location is a difficultask inkeeping up the security of interpersonal organizations. It isbasic to perceive spams in the OSNlocales to spare clients from different sorts of noxious assaults and to protect their security and protection .These unsafe moves received by spammers produces huge network decimation in reality. There are various targets for Twitter spammers likeunconstrained messages, gossipy tidbits,counterfeit news, andspreading invalid data. Spammers accomplish its noxious destinations via notices. In few techniques where diverse mailing records are bolsteredand accordingly dispatch spam messages haphazardly for communicating withdisinclinations. These exercises influence unsettling

(2)

influence to the unique clients who are termedas non-spammers.Likewise,it additionally diminishes OSN stage‟s notoriety.Thusly,itis fundamentalstructure a plan for spotting spammers so that restorative endeavors[30] are taken to counter their malignant exercises.

A few research work shave been done in Twitter spam location space .For including current state- of the-workmanship, couple of overviews are done on counterfeitclient identification from Twitter.Tingminetal give an overview ofnew strategies and procedures to recognize Twitter spam recognition.The above review exhibits a relative report of the present methodologies. [31] Then again, thecreators in ledanoverview on various practices showed by spammers on Twitter informal community. The examination moreovergives a writing audit that perceivesthepresence of spammers on Twitter interpersonal organization. [29] Regardless ofall current considers, there is as yetahole in current writing.Along these lines, for crossing over barrier, we survey best in class in spammer identification as well as phony client identification on Twitter. Additionally, thisreview introducesscientific classification of Twitters pam identification methodologiesand endeavorsfor offeringdefinite portrayal of on goingadvancements in space.

2. Proposedsystem

In this paper, classification of spammer location strategies isexpanded. Fig. 1 shows that the proposed scientific categorization toidentify thespammers on Twitter. Proposed scientific classification is ordered into four primary classes, to be specific,(i)counter itsubstance,(ii)URL based spam discovery,(iii)distinguishing spamininclining subjects, and (iv) counterfeit client identification. Every classification of identification strategies depends ona special model,procedure,and identification calculation.The first class(counterfeit substance)incorporates different procedures, for example, relapse forecast model, malware alarming framework, and Lfunconspire approach. In subsequent classification (URL based spam location),thespammer is identified in URL through various AI calculations. The third classification (spamingdrifting themes)is identified through Naïve Bayes classifier furthermore,language model uniqueness. Last class(counter itclient identification) depends on identifying counterfeit clients through half and half strategies. Procedures identified with everyoneof the spammer identification classes are examined in the accompanying subsections.

A.Fakecontentbasedspammerdetection

Gupta etal. [6] playedout aninside and out portrayal of segments that are influenced by quickly becoming noxious content.It is seen that anenormousindividuals with high socialroles are reliable to circle counterfeit news. To perceive phony records, creatorschoserecordsthat were constructedfollowing Boston impactand are laterrestrictedbyTwitter due to infringement oftermsandconditions.About7.9million particular tweets were gathered by 3.7 million particularclients.Thisdatasetisknown asbiggest Bostonimpact.Dataset. Creatorsplayed outphony substanceorder throughfleeting examination where fleeting appropriationoftweets isdeterminedin lightoftweets quantitypostedeveryhour identificationofphonysubstancewas determined through: normal verified accounts count that wereeither spamornon-spamanddevotees quantity of client accounts. The phonysubstanceproliferationis identified throughmeasurements that includes amiability, social notoriety,themecommitment,worldwide commitment andvalidity.

Fromthatpoint forward,creatorsused relapse forecast model for guaranteeinggeneralindividualseffects whospread phony substancearound then and furthermorefor anticipatingphonycontentdevelopmentinfuture.

Conconeetal. [7]introducedat techniquethat givesharmfulalarmingvia specifiedtweetsset utilization inongoingvanquishedthroughTwitterAPI. Awhile later,tweets

(3)

groupconsideringasimilar subjectissummarized for creatinganalarm. Proposed designisutilized forTwitter posting assessment,perceivingheadwayofpermissibleoccasion,and announcing of that occasionitself. Proposedtechniqueuses datacontainedintweetsat point when a spamor malware is perceived by clients or securityreport is dischargedbycertified specialists. Proposed alarming framework contains following parts:(I) ongoinginformation extractionoftweetsand clients,(ii) filtering frameworkdependentonpreprocessingplan and on NaïveBayes calculationfordisposingtweets with incorrectdata,(iii) informationexamination forspammeridentificationwhere recognition windows are thoroughly canceled by Sigmoid capacityorwhen windowsizearrivesat most extreme,(iv)ready subsystemthat isutilizedwhen occasionis built up, frameworkbunches up tweetsthat are applicable to a similar point where tweetsare recognizedwithbunch barycenterand one that isclosesttobunch place ispicked as entire frameworkgroupdelegate, and (v) criticisminvestigation.The methodologyisprofessed to be efficient and successfulfor discovery of some intrusiveandhonorabledangerousexercises available foruse.

In addition, Eshraqietal. [8] decided different highlightsfor identifyingspamand afterwardusing all airstreambasedbunchingcalculation assistance, spamtweets are perceived.Fewclient accountsare selectedfrom variousdatasetsand shorttime later arbitrary tweetsare selectedfrom theserecords.

Tweetsareinthiswayorderedasspamas well as nonspam. Creators assertedthat computation mayseparate information into spamand non-spam with high precision and phony tweetsperceivedwith highaccuracy as well as exactness.

Different features can be utilized to decide the spams.Formmodel,highlight dependentonchart is statewhere Twitterisformedasdiagram‟s social model.Inaeventthat devoteesquantity is lowinexaminationwithfollowings quantity,recordvalidity islowand likelihood that accountisspamismoderatelyhigh. In likemanner, highlight based on contentincorporatestweets notoriety,HTTPjoins,makesreferencetowhat'smore, answers,and slantingthemes.Forthe timeinclude,numeroustweets aresentby client accountingspecifictimeinterim,atthat pointitisspamaccount. Investigationdataset involvedhas 50,000client accounts.Methodologyidentifiedspammers furthermore,counterittweetswithhighprecision.

A learningfor unlabeled tweets technique, which is used for dealing with different issues in Twitter spam location,is addressed byChen et al.[9].Its structure has two parts,i.e.,gain from recognizedtweets, gain from human marking. Two segmentsareutilized for naturally producing spamtweetsfrom given plaintweetsarrangementthat are handily gathered from Twitter arrangeside.Whennamedspam tweetsare acquired,irregular woodlandcalculationis utilized toperformclassification.The plan exhibition is assessedwhile recognizing floated spam tweets.

Trials areperformed on this present reality informationof tenconsistentdayswith day with100K tweetseachfor spamand nonspam. Discovery rate and F-measure were utilized to assess theexhibition oftheintroducedplot. Theconsequences oftheproposed technique demonstratedthat systemimprovesspam identification‟s precision significantly in this present realitycircumstances.

Moreover,Buntainetal. [10]presenteda strategytodistinguishcounterfeitnews onTwitternaturally via anticipating preciseappraisalintobelievability centereddatasets.Techniqueisappliedon Twitter counterfeitnews datasetand model is prepared againstapubliclysupportedspecialistbased on columnistsevaluation. TwoTwitter datasetsare utilized forexamining respectabilityin OSNs.

CREDBANK, publicly supported dataset, is utilizedfor assessingoccasions exactness in Twitter.

PHEME is awriter nameddataset of conceivablebitsof gossipinTwitter and journalistic assessmentoftheirprecision.An aggregateof45 features wereportrayed that fall into four classifications: basic component, client include, content element, also, transient highlights.

Adjusting marksinPHEME and BUZZFEEDhasclassesthat depict whether rastoryis phony or

(4)

genuine. Consequencesofexamination areuseful in contemplatingdataviaweb- basednetworkingmedia forknowing whethersuchstoriesbolstercomparative example.

B.URL Basedspamdetection

Chen etal. [11] playedout an AI calculationsassessment for distinguishing spam tweets. Creators examined differentfeatures effect on spamidentification‟s exhibition,forinstance:spam to non- spamproportion, preparingdataset size, time relatedinformation, factordiscretization, what'smore, examiningofinformation. Toassess thelocation,first, around 600millionopen tweetswere

gatheredand inthismannerthe creatorsapplied the

Trendminiaturizedscale'swebnotorietyframeworktodistinguishspam tweets howevermuch ascould be expected. An aggregate of 12 lightweight highlights are likewise isolated for recognizing nonspam as well as spamtweetsfromthisidentified dataset.Theidentified highlightsqualities arespokentobyconfigures.

C. Detectingspamintrendingtopic

Ghargeetal.[3] start technique, which is classifiedonpremiseoftwo new viewpoints.Firstoneisspamtweetsacknowledgment with noearlier dataabout clientswhat'smore,subsequentone is language investigation for spamrecognition onTwitterdriftingthemearoundthen. The framework structureincorporates the accompanyingfivesteps.

1. The tweetsassortment regarding driftingpointsonTwitter.Inwake ofputting away tweetsina specificposition,tweets arehenceexamined.

2. Spam labelling isperformed for checking through all datasets that are accessible for recognizing dangerousURL.

3. Feature extractionisolates qualities developing viewof the languagemodel that utilizes language asapparatusand aides in decidingiftweetsare counter feitrnot.

4. The informationalcollection‟s classification isperformedviashortlistingtweetsarrangement that isdepictedbyfeaturesarrangement given toclassifierfor training model andfor securinginformationforspam location.

Staffordet al.[12]analyzedhow much the drifting undertakingsinTwitteraremisused byspammers.

Despitethe factthat varioustechniquestodistinguish the spam areproposed, examination on deciding spam‟s impact on Twitter slanting themeshasaccomplished just constrained considerationof the specialists.The creatorsin[12] introducedprocedureto help out Twitter open API. Actualizedprogramis used to find 10slantingthemesfrom everywhereworld withlanguage code insideone hour and open filtered associationidentified withthosethemes for obtaining information stream.Infollowinghour,creators acquired tosuch an extentofthetweets and connected metadataasallowed bythe Twitter Programming interface. Wheninformationis gathered,gatheredtweetsareclassifiedinto two classifications,i.e.,spamand non-spamtweets, whichareusedforeducatingclassifiers.

For growingsuchmanualmarkingassortment, another programisproposedfortestingirregulartweets, where thought depends on URL filteringby Hussain etal.[20]. Aftermarking tweets consummation,theypushtoward followinginvestigationtechniqueperiod.

There are two separate stages in investigation strategy. Selection and assessment of property is done at the first stage via data recovery measurements. Spam filtering impact on inclining points are accessed using subsequent stages via factual test. The assessment‟sconsequence presumes that spammer doesn't procure inclining theme in Twitter yet on the other hand embraces target points with required characteristics. Outcomes connote well for Twitter supportability and produces a technique for improvement.

(5)

D. Fake Useridentification

AncategorizedstrategyisproposedbyEr³ahinetal. [1] fordistinguishingTwitter spam accounts. The dataset utilizedininvestigationis gathered physically. Classificationis performed viaexaminingclient name,proleand foundationpicture, companions count and devotees, substanceof tweets,depictionofrecord,andtweets count. Datasetincluded501 phonyand 499 genuinerecords, where 16 highlightsfrom datathat were acquired from Twitter APIs are identified. Two examinationsareperformed to characterizecounterfeit records. Firsttestutilizes Naïve Bayes learning calculationonTwitterdataset which includes allangleswithout discretization,thoughsubsequent analysisemploymentsNaïveBayes learning calculation on Twitter datasetafter data is discretized.

Mateenetal.[13] proposed ahalf andhalf procedurethat usesclient ,content,and chart basedqualitiesforspammerprolesrecognition. Amodel isproposedfor separating betweennon- spamand spamprofilesutilizingthree attributes.Proposedprocedureis broke downutilizing Twitterdatasetwith11Kclientsand 400Ktweets.Objectivesare accomplishinghigher efficiencyandaccuracyviacoordinatingeveryoneofthesequalities. Client based highlightsare setupinrelationshipview andclientaccountsproperties. Itis fundamental toadd clientbased highlights for spamrecognitionmodel. Asthese highlightsare identified with clientaccounts, all characteristics, which are connectedto client accounts,are identified.Thesepropertiesincorporatedevoteesquantity what's more,after,age,FFproportion,and notoriety.Ontheother hand,content highlightsareconnectedtotweetsthat are postedbyclientsas spambotsthat post colossalmeasure ofcopy substance asdifferenceto non-spammerswho don't postcopytweets.

Thesefeaturesrelyupon substanceor messages thatclientscompose.Spammerspresentsubstanceon spread phonynews and these substance includes noxious URLto advance their item.

Substancebased highlights includes: all outnumber of tweets, hash tag proportion,URLsproportion,specifiesproportion,and frequencyoftweets.Chart basedelementisusedforcontrolling avoidance techniques that are led by spammers. Spammers utilizevariousmethods forabstainingfrombeing distinguished.They canpurchase counterfeit adherents fromvarious thirdpartysites andtrade theirdevoteesto anotherclient to resemble legitimate client.Chartbasedhighlightsrememberforoutdegree and betweenness. The methodology‟sassessment isdone byutilizing past methods dataset as, due to Twitter approach, no information is accessible openly. The outcomesareassessed via coordinatingthree mostnormal methodologies,specifically Decorate,NaïveBayes,and J48. The aftereffectoftest shows that recognitionpaceofmethodologyismuch precise andhigherthananyofcurrentmethods.

Guptaetal.[14]presentedastrategyforspammers discovery inTwitterandutilized mainstreamprocedures,i.eNB(Naïve Bayes), bunching, and choicetrees.Calculationsgroup a recordas spamornon-spam.Dataset includes1064 Twitterclients which includes62 highlights, that are either specific to tweet or client data.Spammeraccounthaspractically36%ofutilizeddataset.

Asspammersconductisnot sameasnon- spammers, afew characteristics or highlightsare perceivedinwhichthetwoclassificationsarenotquite the sameas eachother. Highlight identification isbasedonhighlightsthe quantity atclient and tweet level,forexample,devotees orfollowing,spam watchwords,answers, hash tags,andURLs[30],[32].

After features the identification, pre-processor stepchangesevery singleconsistentcomponentinto discrete.Therefore,thecreators builtupasystem utilizinggrouping,choice trees,NaïveBayes calculations.WithNaïveBayes,records were identified via assessment ofcertainrecordchance as non-spammer orspammer.Inbunchingbased calculation, whole records arrangement is classified intotwoclassesasnon-spamor spam.

(6)

3. Conclusion

In thispaper, various strategies utilized for identifyingspammersontwitterdata set were analyzed.

Spammers were identified based onvarious techniques like URL used by the user, trending topics and fake content shared by the users and fake users. Different kinds of features like user id, Retweets count, likes count etc which are posted by the users are identified from which the nature of the user is detected as spam or not. In future, it can be extended with various advanced machine learning algorithms for identifying spammers.

References

[1] B.Erçahin,Ö.Akta,D.KilinçandC.Akyol, „Twitterfakeaccount detection‟,inProc.

Int.Conf.Comput.Sci.Eng.(UBMK),Oct.2017, pp.388_392.

[2] F.Benevenuto,G.Magno,T.Rodrigues,andV. Almeida detecting spammerson Twitter,‟

inProc. Collaboration,Electron. Messaging, Anti-

AbuseSpamConf.(CEAS),vol.6,Jul.2010,p.12.

[3] S.Gharge,andM.Chavan, „Anintegratedapproachformalicioustweets detectionusingNLP‟,inProc. Int.Conf.InventiveCommun. Comput.Technol.

(ICICCT),Mar.2017,pp.435_438.

[4] T.Wu,S.Wen,Y.Xiang,and W.Zhou, „Twitterspamdetection:

Surveyofnewapproachesandcomparative study,‟ Comput.Secur., vol.76,pp.265_284,Jul.2018.

[5] S.J.Soman, „Asurveyonbehaviorsex habited

byspammersinpopularsocialmedianetworks‟,inProc.

Int.Conf.Circuit,PowerComput.Tech-no. l.(ICCPCT),Mar.2016,pp.1_6.

[6] A.Gupta, H. Lamba, and P. Kumaraguru, 1.00 per RT#BostonMarathon#prayforboston:Analyzing fake content onTwitter‟, inProc.

eCrimeResearchersSummit(eCRS),2013,pp.

[7] F.Concone,A.DePaola, G.LoRe, and M.Morana, „Twitter analysisforreal-timemalware discovery‟, inProc. AEITInt.Annu. Conf.,Sep.2017,pp.1_6.

[8] N.Eshraqi,M.Jalali, and M.H.Moattar, „Detecting

spamtweetsinTwitterusingadatastreamclustering algorithm‟, inProc. Int.

Congr.Technol.,Commun.Knowl.(ICTCK),Nov.2015,pp.347_351.

[9] C.Chen,Y.Wang,J.Zhang,Y. Xiang,W.Zhou,andG.Min, „Statistical features-based real-

time detection of drifted Twitter spam‟,

IEEETrans.Inf.ForensicsSecurity,vol.12,no.4,pp.914_925, Apr. 2017.

[10] C.BuntainandJ.Golbeck,

„AutomaticallyidentifyingfakenewsinpopularTwitterthreads‟,inProc.IEEEInt.

Conf.SmartCloud (SmartCloud),Nov.2017,pp.208_215.

[11] C.Chen,J.Zhang,Y.Xie,Y.Xiang,W.Zhou,M.M.Hassan,A.AlElaiwi,and M.Alrubaian,

„A performanceevaluationofmachinelearning-

basedstreamingspamtweetsdetection‟,IEEETrans.

Comput.SocialSyst.,vol.2,no.3,pp.65_76,Sep.2015.

(7)

[12] G.StaffordandL.L.Yu, „An evaluation oftheeffectofspamonTwittertrending topics,‟

inProc.Int.Conf.SocialComput.,Sep.2013, pp.373_378.

[13] A. Gupta andR.Kaushal, „Improvingspamdetectioninonlinesocial networks‟,in Proc.

Int. Conf. Cogn. Comput.Inf. Process. (CCIP),Mar.2015,pp.1_6.

[14] F.Fathalianiand M.Bouguessa, „Amodel-basedapproachforidentifying

spammersinsocialnetworks‟, inProc. IEEE Int.Conf.

DataSci.Adv.Anal.(DSAA),Oct.2015,pp.1_9.

[15] V.Chauhan,A.Pilaniya,V. Middha,A.Gupta,U.Bana,B.R. Prasad,andS. Agarwal,

„Anomalous behaviordetectionin social networking‟' in Proc. 8thInt.Conf. Comput., Commun. Netw. Technol. (ICCCNT),Jul.2017,pp.1_5.

[16] S.Jeong,G.Noh,H.Oh,andC.-K.Kim, „Followspamdetectionbased on cascaded socialinformation‟,Inf.Sci.,vol. 369,pp.481_499,Nov.2016.

[17] M.Washha,A. Qaroush, and F.Sedes, „Leveraging time for spammersdetection on Twitter‟, in Proc. 8thInt.Conf. Manage. Digit.EcoSyst.,Nov.2016,pp.109_116.

[18] B.Wang,A.Zubiaga,M.Liakata,andR.Procter, „Makingthe most of tweet-inherent features for social spam detection on Twitter,‟ 2015,arXiv:1503.07405. [Online].

Available: https://arxiv.org/abs/1503.07405

[19] M.Hussain,M.Ahmed,H.A.Khattak,M.Imran,A.Khan,S.Din, A.Ahmad, G.Jeon, andA.G. Reddy, „Towards ontology-based multilingual URL_ltering:Abigdataproblem‟, J.Supercomput.,vol.74,no.10,pp.5003_5021,Oct.2018.

[20] C.Meda,E.Ragusa,C.Gianoglio,R.Zunino,A. Ottaviano,E. Scillia, andR. Surlinelli,

„Spam detectionof Twitter traf_c: A frameworkbasedon random forests and non-

uniform feature

sampling‟,inProc.IEEE/ACMInt.Conf.Adv.SocialNetw.Anal.Mining(ASONAM),Aug.

2016,pp._817.

[21] S.Ghosh,G.Korlam,andN.Ganguly,

„Spammers'networkswithinonlinesocialnetworks:Acase-studyonTwitter‟, inProc. 20th Int.Conf.CompanionWorldWideWeb,Mar.2011,pp.41_42.

[22] C.Chen,S.Wen,J.Zhang,Y.Xiang,J.Oliver,A.Alelaiwi,andM. M. Hassan,

„InvestigatingthedeceptiveinformationinTwitter spam‟,

FutureGener.Comput.Syst.,vol.72,pp.319_326,Jul.2017.

[23] I.David,O.S. Siordia,andD. Moctezuma,`Features combinationforthe detectionofmaliciousTwitteraccounts‟, in Proc. IEEE Int.Autumn MeetingPower,Electron. Comput.(ROPEC),Nov.2016,pp.1_6.

[24] M.Babcock,R.A.V. Cox,andS.Kumar, „Diffusionofpro-and anti-false information

tweets: The black panther movie case‟,

Comput.Math.Org.Theory,vol.25,no.1,pp.72_84,Mar.2019.

[25] S.Keretna, A.Hossny, and D.Creighton, „Recognisinguseridentity

inTwittersocialnetworksvia text mining‟,inProc.

IEEEInt.Conf.Syst.,Man,Cybern.,Oct.2013,pp.3079_3082.

[26] C.Meda, F.Bisio, P.Gastaldo, andR. Zunino, „Amachinelearning approachforTwitter

(8)

spammersdetection‟, in Proc.Int.CarnahanConf.Secur. Technol.

(ICCST),Oct.2014,pp.1_6.

[27] SenthamilSelvi.R and Valarmathi ML, „An improved firefly heuristics for efficient feature selection and its application in big data‟ in International journal of Biomedical Research, Vol 1, Issue 1, Feb 2017, pp. 236-241.

[28] S Mohana and Dr.S.A.Sahaaya Arul Mary, "Privacy preserving in Health Care Information:AMemetic Approach", Journal of Medical Imaging & Health Informatics‟,American Scientific Publishers, pp. 779-783, Vol. 6, Issue. 3, 2016.

[29] T.M.Nithya, J. Ramya, L. Amudha,“Scope Prediction Utilizing Support Vector Machine for Career Opportunities”, International Journal of Engineering and Advanced Technology (IJEAT), ISSN: 2249- 8958, Volume-8 Issue-5, June 2019, pp.2759-2762.

[30] L. Amudha, Dr.R.PushpaLakshmi, “Scalable and Reliable Deep Learning Model to Handle Real-Time Streaming Data”, International Journal of Engineering and Advanced Technology, ISSN: 2249 – 8958, Volume-9 Issue-3, February, DOI:

[31] T.M.Nithya, K.S.Guruprakash, L.Amudha. (2020). DEEP LEARNING BASED PREDICTION MODEL FOR COURSE REGISTRATION SYSTEM. International Journal of Advanced Science and Technology, 29(7s), 2178-2184

[32] Nithya, T.M., Chitra, S.. (2020). Soft computing-based semi-automated test case selection using gradient-based techniques. Soft Computing. 24. 12981–12987 (2020) [33] K.S.Guruprakash, R.Ramesh, Abinaya K, Libereta A, Lisa Evanjiline L, Madhumitha

B. (2020). Optimized Workload Assigning System Using Particle Swarm Optimization. International Journal of Advanced Science and Technology, 29(7), 27.