Classification Algorithm in Machine Learning.pdf

Classification Algorithm in Machine Learning.pdf

1. Classification Algorithm in
Machine Learning

2. Classification Algorithm in Machine Learning
•TheSupervisedMachineLearningalgorithmcanbebroadlyclassified
intoRegressionandClassificationAlgorithms.
•InRegressionalgorithms,wehavepredictedtheoutputforcontinuous
values,
•Topredictthecategoricalvalues,weneedClassificationalgorithms.

3. Classification Algorithm
•TheClassificationalgorithmisaSupervisedLearningtechniquethatis
usedtoidentifythecategoryofnewobservationsonthebasisof
trainingdata.
•InClassification,aprogramlearnsfromthegivendatasetor
observationsandthenclassifiesnewobservationintoanumberof
classesorgroups.
•YesorNo,0or1,SpamorNotSpam,catordog,etc.Classescanbe
calledastargets/labelsorcategories.

4. •Unlikeregression,theoutputvariableofClassificationisacategory,
notavalue,suchas"GreenorBlue","fruitoranimal",etc.
•SincetheClassificationalgorithmisaSupervisedlearningtechnique,
henceittakeslabeledinputdata,whichmeansitcontainsinputwith
thecorrespondingoutput.
•ThebestexampleofanMLclassificationalgorithmisEmailSpam
Detector.

6. Types of classifiers
•BinaryClassifier:Iftheclassificationproblemhasonlytwopossible
outcomes,thenitiscalledasBinaryClassifier.
Examples:YESorNO,MALEorFEMALE,SPAMorNOTSPAM,CATor
DOG,etc.
•Multi-classClassifier:Ifaclassificationproblemhasmorethantwo
outcomes,thenitiscalledasMulti-classClassifier.
•Example:Classificationsoftypesofcrops,Classificationoftypesof
music.

7. Types of ML Classification Algorithms:
•Linear Models
•Logistic Regression
•Support Vector Machines
•Non-linear Models
•K-Nearest Neighbours
•Kernel SVM
•Naïve Bayes
•Decision Tree Classification
•Random Forest Classification

8. Logistic Regression

10. Logistic Regression
•LogisticregressionisoneofthemostpopularMachineLearning
algorithms,whichcomesundertheSupervisedLearningtechnique.
•Itisusedforpredictingthecategoricaldependentvariableusinga
givensetofindependentvariables.
•Logisticregressionpredictstheoutputofacategoricaldependent
variable.Thereforetheoutcomemustbeacategoricalordiscrete
value.
•ItcanbeeitherYesorNo,0or1,trueorFalse,etc.butinsteadof
givingtheexactvalueas0and1,itgivestheprobabilisticvalues
whichliebetween0and1.

11. •LogisticRegressionismuchsimilartotheLinearRegressionexcept
thathowtheyareused.
•LinearRegressionisusedforsolvingRegressionproblems,
whereasLogisticregressionisusedforsolvingtheclassification
problems.
•InLogisticregression,insteadoffittingaregressionline,wefitan"S"
shapedlogisticfunction,whichpredictstwomaximumvalues(0or1).
•Thecurvefromthelogisticfunctionindicatesthelikelihoodof
somethingsuchaswhetherthecellsarecancerousornot,amouseis
obeseornotbasedonitsweight,etc.

16. Cost function: MSE is not used

19. Logistic Function (Sigmoid Function):
•Thesigmoidfunctionisamathematicalfunctionusedtomapthe
predictedvaluestoprobabilities.
•Itmapsanyrealvalueintoanothervaluewithinarangeof0and1.
•Thevalueofthelogisticregressionmustbebetween0and1,which
cannotgobeyondthislimit,soitformsacurvelikethe"S"form.The
S-formcurveiscalledtheSigmoidfunctionorthelogisticfunction.
•Inlogisticregression,weusetheconceptofthethresholdvalue,which
definestheprobabilityofeither0or1.Suchasvaluesabovethe
thresholdvaluetendsto1,andavaluebelowthethresholdvalues
tendsto0.

20. Type of Logistic Regression:
Onthebasisofthecategories,LogisticRegressioncanbeclassifiedinto
threetypes:
•Binomial:InbinomialLogisticregression,therecanbeonlytwo
possibletypesofthedependentvariables,suchas0or1,PassorFail,
etc.
•Multinomial:InmultinomialLogisticregression,therecanbe3or
morepossibleunorderedtypesofthedependentvariable,suchas
"cat","dogs",or"sheep"
•Ordinal:InordinalLogisticregression,therecanbe3ormore
possibleorderedtypesofdependentvariables,suchas"low",
"Medium",or"High".