AI Builder - Binary Classification
Download as DOCX, PDF0 likes149 views
This document provides instructions for building and deploying a binary classification model to predict online shopper purchase intentions using historical data. The model is built and trained using an online shopper data set containing site visit details and sales revenue. Model accuracy is evaluated and details on influential data fields are reviewed. Finally, a simple model-driven Power Apps application is created to view the prediction data and scores. The application displays predicted revenue, associated probabilities, and actual revenue for analysis.
AI Builder - Binary Classification
- 1. Binary classification Binaryclassificationuseshistorical datatopredictwhethernew datafallsintoone of twocategories. You can use these insightstodevelopactionable intelligence foryourbusiness. In thislab,we will buildandtraina binaryclassificationmodel. We will alsoreview the level of influence of the selectedcriteria. Finally,we will make amodel-drivenapplicationto review thisdataprediction. The data usedhere will helpuspredictonlineshopperintentions. We have detailsof ashopper’ssite visitaswell asif thatvisitresultedinasale (revenue). Once we have the predictioninformation,we can use that to helpguide marketingplans,websiteupdates,promotional communications,andmore. Note: If you are buildingthe firstmodel inanenvironment,clickonExplore Templatestogetstarted. Exercise 1 In the firstexercise youwillbuildandtrainyourmodel. 1. From the leftnavigation,expandAIBuilderandselectBuild.
- 2. 2. SelectBinaryClassification. 3. Name yourmodel. Because youare workingina sharedenvironmentmake sure toinclude your name as part of the model name. Thiswill make iteasiertofindlater. Clickcreate.
- 3. 4. Your screenshouldlooklike the followingimage. 5. Notice the progressindicatoronthe left.
- 4. 6. You will see Quicktipsonthe right. 7. In the centerwe will buildourmodel. Clickinthe EntityfieldandselectOnline Shopper Intention. Thisisacustomentitythat collectsthe datawe are storingregardingthe online shopper’svisittooursite.
- 5. 8. From the selectedentitywe willnow see availablebinaryfieldsforprediction. The onlyfields we will see here are of type “Twooption.” SelectRevenue.ClickNext. 9. The nextscreenwill showusavailable fieldsonthe entity. Eachfieldselectedwill be evaluated by the model forthe fieldsinfluence onthe endresult. Bydefault,all fieldswill be selected,do not remove anyfields. ClickNext. 10. Notice our progress indicatorhasmovedtothe nextitem. ClickTrain.
- 6. 11. Your Binary Classificationmodel will now train. Goto Models. 12. Locate andopenyour savedmodel. If youneedhelpfindingit,type yourname intothe search box. 13. Aftereachtraining,AI Builderusesthe testdatasetto evaluate the qualityandaccuracyof the newmodel.A summarypage foryour model showsyourmodel trainingresult,includinga Performance score.The score we have is 65% and we shouldexpectittochange overtime. And now we can viewsome detailsaboutourtrainedmodel. ClickonView Details.
- 7. 14. Our detailsreportwill show the datainfluencersandhow muchinfluencethatdatahas on our predictionmodel. Close the details. 15. Publishyourmodel. Once publishedthe datawill getscored,andscoringwill happendailyfora publishedmodel.
- 8. About that score: Performance score calculations AI Buildercalculatesthe performance score foryourmodel basedonthe precisionandrecall of the predictionresults: Performance score: Thisis the harmonicmeanof precisionandrecall.Itbalancesbothfor an imbalancedclassdistribution.Performance score valuesare between0 - 100. Generally,the higherthe performance score,the betteryourmodel performs. Precision:The fractionof correct predictionsamongall the positive predictions. Recall: The fractionof correct predictionsamongall true positive cases. Exercise 2 We will make asmall model drivenapplicationtoview the data. Thisapproachallowsusa quicklookat the data. You couldalsomake a canvasapp that allowsuserstointeractwiththe data froma mobile device. For thepurposesof this lab,we havetaken someshortcutsin the interest of time. In general,you should alwaysbeworking in a specific solution,you should rename items with smartnamesforbetter teamdevelopmentand more. A full lesson on customization and solution strategy isbeyond thescopeof theselabs. However, Microsofthasmany learning choicesavailable. Please askyourinstructorif you’d like moreinformation. 1. In yourPowerAppsmakerportal,fromthe leftside navigation,navigatetoAppsandselectCreate an app and selectModel-driven.
- 9. 2. If this isyourfirsttime connectingtoa canvas app inthisenvironment,youmightbe promptedto choose yourregion. Selectthe default. 3. Give yourapp a name,remembertoinclude yourownname aspart of it. Leave all otheroptionsas theyare andclick Done. 4. You shouldsee somethinglikethe image below.
- 10. 5. We will nowaddouronline shopperintention entitytothe site map. Clickthe editicononthe Site Map. 6. Withfocus seton the Areagive the area a name such as Online ShopperIntentions. Selectand name the Group as well.
- 11. 7. Withthe NewSubareaselected selectType of EntityandOnline ShopperIntentionasthe entity. Leave otherfieldsasdefaults. 8. Save your Site Map and go backto the AppDesigner. 9. You shouldnowsee the assetsneededforthe OnlineShopperIntentionentityare includedinour model-drivenapp. 10. Let’sadd a viewtoour app so we can bettersee the predictiondata. ClickonViewsandthenCreate New.
- 12. 11. If promptedwithunsavedchanges,youcandismiss byclickingOK (assumingyou’ve followedour stepsso far!). 12. ClickColumnAttributesPrimaryEntity. 13. Drag and drop the followingfieldstothe view: revenue-predicted(modelname);revenue- probability(modelname);andrevenue. Note:You will see fields added forthemodelsof all of thestudentsworking in the same environment,lookfortheoneswith yourmodelnamein thefield name.
- 13. 14. Adjustcolumnwidthstomake themwiderandeasiertoview. With the columnselected,increase the widthusingthe optionsonthe rightside of the view designer. 15. Save your view,remembertoinclude yourname aspart of the view name.
- 14. 16. Returnto AppDesigner. Save yourappand publishit. 17. Playyour app. 18. You shouldnowbe inyour newmodel-drivenapp. Selectthe view youjustcreated.
- 15. 19. Viewthe data. You can see the predictionscore,andthe predictionthatfollows. Withthis informationyoucancreate automationsuchas a flow that wouldtrigger basedona customer’s expectedrevenueandsenddiscountemailstocustomerslikelytocompleteapurchase. Creating such automationisbeyondthe scope of thisworkshop.