PGP-DS Capstone Project Industry Review Writing

Guidelines for PGP-DS Capstone ProjectIndustryReview

• IndustryReviewCurrentPractices,BackgroundResearch
• LiteratureSurvey-Publications,Applications,pastandundergoingresearch

DatasetandDomain

• data dictionary
• Variable categorization (countofnumericand categorical)
• Pre-ProcessingDataAnalysis(countofmissing/nullvalues,redundantcolumns,)
• Alternatesourcesofdatathatcansupplementthecoredataset(atleast 2-3 columns)
• Project Justification -Project Statement, Complexity involved, Project Outcome Commercial,AcademicorSocialvalue

data exploration(EDA)

• Relationship between variables
• Check for
• Multi-colinearity
• Distribution of variables
• Presence of outliersandtheirtreatment
• Statistical significance ofvariables
• Class imbalance anditstreatment

FeatureEngineering

• Whether any transformationsrequired
• Scaling the data
• Feature selection
• Dimensionality reduction

Assumptions

• Check for theassumptionsto besatisfiedforeachofthemodelsin
• RegressionSLR,MultipleLinearRegression,LogisticRegression
• ClassificationDecisionTree,RandomForest,SVM,Baggedandboostedmodels
• Clustering PCA (multi-co linearity), K-Means (presence of outliers, scaling, conversion to numerical,)

InterimPresentationCheckpoint

Modelbuilding

• Split the datatotrainand test.
• Start with asimplemodelwhichsatisfiesalltheaboveassumptionsbasedonyour
• Check for biasandvariance
• To improve the performance, try cross-validation, ensemble models, hyperparameter tuning,gridsearch

Evaluationofmodel

• RegressionRMSE,R-Squaredvalue,
• ClassificationClassificationreportwithprecision,recall,F1-score,Support,AUC,
• ClusteringInertiavalue
• Comparison of differentmodelsbuiltanddiscussionof thesame
• Time taken fortheinferences/predictions

BusinessRecommendations&Futureenhancements

• How to improvedatacollection,processing,andmodelaccuracy?
• Commercial value/Socialvalue/Researchvalue
• Recommendations based oninsights

FinalPresentationCheckpoint

Dashboard

• EDACorrelationmatrix,pairplots,boxblots,distributionplots
• Model
• ModelParameters
• Visualization of performanceof themodelwithvaryingparameters
• Visualization of modelMetrics
• Testing outcome
• Failure cases andexplanationforthesame
• Most successful andobviouscases
• Border cases

FinalSubmissionCheckpoint