38
Neural Networks Artificial Neural Networks (ANN) Process of Machine Learning Directed/Supervised Data Mining Applications in “Prediction” Fraud Detection Customer Response Credit Rating

Session 5 - Neural Networks

Embed Size (px)

DESCRIPTION

Session 5 - Neural Networks

Citation preview

Neural NetworksArtifcial Neural Networks (ANN)Process of Machine LearningDirected/Supervised Data MiningApplications in Prediction!"raud Detection#usto$er %esponse#redit %atingNeural NetworksMi$ic neurons of the hu$an &rainLinks are the Processing 'le$entsLearn fro$ e(perience)ood in detecting unknown relationshipsP's process data &* su$$ari+ing and transfor$ing it through $athe$atical functionsNeural NetworksP's are interconnected and trained and retrained repeatedl*P's are linked to inputs and outputs,raining involves $odif*ing the weight or connection-ses Learning %ules! to ad.ust weights,raining continues till desired accurac* level is reachedNeural Network ModelAgeRegionCall RateServiceIncomeLoyalHopperLostSi$ple Network1235/0446/1W14W24W25W15W34W35W46W562aria&le Description34 5nput 2aria&les (Age6 5nco$e 'tc)Nu$eric or #ategorical7(i6.)4 7eights associated with each P'#onverts the input values to a nodal value,heta (i)4 8ias at each node9riginal 2aluesriginal Wts! riginal Wts!"1 1 W34 #$!5"2 $ W35 $!2"3 1 W46 #$!3W14 $!2 W56 #$!2W15 #$!3 %&eta4 #$!4W24 $!4 %&eta5 $!2W25 $!1 %&eta6 $!1Si$ple Network123546$!2$!4$!1#$!3#$!5$!2#$!3#$!2 #lass 2alue :;?=>@=>;9utput at Node ?5nput to Node ?3;:;A3@:=A 3B:;7;?:=>@A 7@?:=>?A 7B?:0,heta(?) : ?Net 5nput to Node ?: (;)(=>@)C(=)(=>?)C(;)(0)C(?):D9utput E node ? :;/(;C;/eBB@-se SFuashing "unction!9utput at Nodes 0 and 15nput to Node 03;:;A 3@:=A3B:;7;0:BA 7@0:=>;A7B0:=>@,heta(0) : =>@Net 5nput to Node 0: (;)(B)C(=)(=>;)C(;)(=>@)C(=>@):=>;9utput E node 0 :;/(;C;/e;):=>0@0Si$ilarl* at Node 1 :=>?D?#alculation of 'rror at Nodes#alculated 2alue at Node 1:91:=>?D?Actual 2alue :,1:;>='rror :'1: (91)(;;B;;'rror at 0 :(90)(;BB@)(;BB@)(B)(=>;B;;):==GD#alculation of 'rror of 7eights'rror of 7?1:(L)(9?)H('1):(=>I)(=>BB@)(=>;B;;):=>=BIL is the rate of learning! (&etween = and ;)%evised 2alue of 7?1 : 9ld 2alue C 'rrorNew 7?1:BC=>=BI:@1;'rror of 701:(=>I)(=>0@0)(=>;B;;):=>=1@New 701:@C=>=1@:;BGNew 2aluesBuilding Predictive Models for Election Results in India An Application of Classification Trees and Neural Networks@==? 'lection Predictions@==? 'lections Jarnataka StateParty 11th Assembly 12th AssemblyIndian NationalCongress 133 66Bharatiya Janata Party 43 79Janata Dal !" 1# $%Janata Dal &" 1% $Inde'endents 19 12(thers 1 4)otal 224 224Special "eatures of @==?'lectronic 2oting Machines?== $illion voters%esults announced in G hours'(pectations turned upside down5nfor$ation on the #andidate (Supre$e #ourt K @==@)Criminal o**en+e in the 'ast,hether the +andidate is a++-sed o* any 'ending +ase o* any o**en+e )he assets o* the +andidate b-t also o* his.her s'o-se and the de'endents/0iabilities1 i* any1 'arti+-larly to any '-bli+ *inan+ial instit-tions or go2ernment3d-+ational 4-ali*i+ations o* the +andidateA$end$ent to the Act-nani$it* a$ong @; Political Political partiesA$end$ent to %epresentation of People Act!8rought out as an ordinanceStruck down &* Supre$e #ourt in @==B9&.ectives of the Stud*De2elo''redi+ti2emodels5hi+h+o-ldbe -sed*or'redi+tingtheo-t+omeso*the ele+tionIdenti*y2ario-sas'e+tso*in*ormationmade a2ailable+onse4-enttothe!-'remeCo-rt 6-dgmentandthes-bse4-entorderso*the 3le+tion Commission and32al-ate the relati2e im'ortan+e o* these as'e+ts o* in*ormation in 'redi+ting the ele+tion o-t+omes/ 2aria&les -sedAge o* the +andidate binned" N-mber o* +ontestants Binned"7o2able assets binned into 3 +ategories"Immo2able assets binned into 3 +ategories")otal Assets binned into 3 +ategories"0iabilities binned into 3 +ategories"(5nershi' o* +ommer+ial b-ildings binned"(5nershi' o* residential b-ildings binned"2aria&les -sed,hether the +andidate belongs to the r-ling 'arty or not8e2en-e Di2ision o* the state 9eogra'hi+al area,hether the +onstit-en+y 5as reser2ed )y'e o* 'oliti+al 'arty binned",hether the +andidate is an in+-mbent,hether the +andidate belongs to the in+-mbent 'arty :ariables &sed9ender3d-+ational le2el binned",hether the +andidate had any +riminal re+ord,hether the +andidates o5ns any agri+-lt-ral land,hether the +andidate has any liabilities to *inan+ial instit-tions,hether the +andidate has any liabilities to go2ernment#andidate ProfleCategory Freuency Percent!nknown "#" $%&'(Pri)ary *c+ool ,- "&%".ig+ *c+ool -'- "$&'$Pre/!niversity "-$ $,&0(1raduate ,,# "#&",Post 1raduate "2, $#&2"Fe)ale (% '&",Male $''' 2,%3oes not 4elong $,%2 (2&'"Belongs $#" $0&,(Not incu)4ent $,20 20&(Incu)4ent Candidate $'$ 2&"!nknown -(# "-&'(B5P $## $0Congress $2% $$&2,6t+er National Party "$0 $"&(53 7*8 $2- $$%6t+er Regional Party "2' $#&2(Independent $(- $$&$'Education1enderBelongs to Incu)4ent Party In t+e constituencyIncu)4entType of Party#andidate ProfleCategory Freuency Percent3oes Not 6wn '2" -%&0(6wns '20 -'&2'!nknown ,'2 "#&2#3oes Not 6wn %00 -%&'%6wns '#- -,&2"!nknown ,%( "(&'"3oes not Belong $,,' ((&0%Belongs $2% $$&2,9ess t+an -0 $$0 %-0 to ,0 ,"( "%&0(,0 to '0 '0# -0&2'0 to %0 ,0' ",&%(%0 to #0 $,% (&2More t+an #0 "" $&-,!nknown ,-# "%&%-No 3ues to 1overn)ent $$,0 %2&,#6wes dues to 1overn)ent %, -&26wners+ip of .ouses6wners+ip of Co))ercial BuildingsBelongs to Ruling PartyAge.as 1overn)ent 3ues#andidate ProfleCategory Freuency Percent!nknown '"% -"&0'No 3ues to FIs 2$# ''&((6wes 3ues to FIs $2( $"&0#!nknown -2% ",&$-No 3ues to Banks '-" -"&,"6wes 3ues to Banks #$- ,-&,'!nknown -#2 "-&$No Agricultural 9and -%( ""&,-6wns Agricultural 9and (2, ',&,(3oes not +ave Cri)inal Record $,'' ((&%#.as Cri)inal Record $(% $$&--:;% ,"# "%&0"# to 2 ,'# "#&('$0 to $" -,, "0&2%< $" ,$- "'&$#3oes not 4elong $-2$ (,#Belongs "'0 $'&"-.as 3ues to Financial Institutions.as 3ues to Banks6wns Agricultural 9andCri)inal RecordNo& of Candidates in t+e ConstituencyReserved ConstituencyMethodolog*-sed #lassifcation ,rees and Artifcial Neural NetworksPro&le$ of Skewed Distri&utionApplied 9verIs 4/3In+-mbent Party 3(riginal Di2isions 3/%9ender 1/38esidential 'ro'erty 2/90iabilities to 9o2t/ $/1 In+-mbent 1/7 8eser2ed 1/2Immo2able Assets 2/$)otal 0iabilities $/3 Party )y'e 198e2en-e Di2isions 4/97o2able Assets 3/68-ling Party 1/9)otal Assets 4/7(thersDemogra'hi+ Chara+teristi+s (5nershi' 0iabilities Politi+al >a+tors#onclusions)he !-'reme Co-rt 6-dgment = dis+los-res o* the ba+