際際滷

際際滷Share a Scribd company logo

Predictive Analytics for Cars.com

Chris Zhu
Chris Zhu

This document summarizes a project using predictive analytics for Cars.com. It describes using k-means clustering and decision tree models on vehicle data to understand customer purchase behaviors and ratings. Key findings include higher ratings being associated with positive dealer service and buying processes. The document proposes updates to Cars.com like additional review questions and ratings to provide better insights for customers. It aims to help customers find dealers that match their needs and preferences.

1 of 33
+ Predictive Analytics for Cars.com Directed by Professor Martin Bariff Xiaofei Yang, Joyce Yi Shi, Ivy Yingqiu Zhu, Chris
Introduction Data Preparation + Data Mining Methods Better Cars.com 2
+ Introduction Yingqiu Zhu, Chris 3
+ 4 INTRODUCTION SPSS MODELER
+ Data Preparation Yingqiu Zhu, Chris
+ 6 DATA PREPARATION KEY VARIABLES a. Basic Information b. Ratings c. Indicators
+ 7 DATA PREPARATION CLEANING Shopping Quality for Used of Repair Overall Rating Ready-to-Use Data
+ 8 DATA PREPARATION BALANCING Unbalanced Data Balanced Data 16.43% 30.21% 3.01% 5.54% 2.11% 3.88% 10.01% 18.40% 68.45% 41.96% 27 鐚
+ Data Mining Methods Yingqiu Zhu, Chris & Xiaofei Yang, Joyce
+ 10 METHODS K-Means Cluster Analysis Decision Tree Predictive Analysis Text Mining
+ 11 CLUSTER ANALYSIS K-MEANS K-Means is the most popular nonhierarchical clustering method. K-Means is based on geometric notions of similarity
+ 12 CLUSTER ANALYSIS K-MEANS Model Summary Model Accuracy
R13 ANALYSI S Yes No 5 1 Importance 100% K-Means 5 1 5 1 + Yes No Importance 11%
+ 14 CLUSTER ANALYSIS K-MEANS Customers who purchase used cars from a dealer rating high for the dealer recommending the dealer to others.
+ 15 PREDICTIVE ANALYSIS INTRODUCTION OF DECISION TREE MODEL A tree-building algorithm Recursively splitting the data into smaller and smaller groups based on the fields that provides the maximum information gain Used C5.0 model in our research
+ 16 PREDICTIVE ANALYSIS DECISION TREE Decision Tree with Balanced Data 60% 40% Evaluation Standard Overall accuracy: > 80% Favorability between training & testing samples: The results with the testing sample compare favorably to the training sample.
+ 17 PREDICTIVE ANALYSIS DECISION TREE Rating - Customer Service
+ 18 PREDICTIVE ANALYSIS INTRODUCTION OF TEXT MINING MODEL Extract key concepts from the text and create categories with these concepts by using linguistic and frequency techniques Explore more information and context included in the data Better compare the results with and without text data and see the difference
+ 19 PREDICTIVE ANALYSIS DECISION TREE MODEL WITH TEXT DATA Model with Only Numeric Data Model with Text & Numeric Data
+ 20 PREDICTIVE ANALYSIS Interactive Workbench Categories and Concepts View
21 Trustful & Knowledgeable Extended warranty Relatively consistent pricing buying/buying process Nearest location < Service & Credibility
+ 22 PREDICTIVE ANALYSIS TEXT MINING Rating = 5 Rating = 1 High Overall Rating Low Overall Rating 100 90 100 80 90 80 70 70 60 60 50 50 40 40 30 30 20 20 10 10 0 0 % %
+ Better Cars.Com Yi Shi (Ivy)
What we get from our analysis 24 Less influential dealers Purchase behavior information =>Higher overall rating Overall Rating Higher rating for Text Mining buying process & customer service =>New Hint for Consumer behavior =>Higher Overall Rating
+ 25 Stimulus Influence Overall Rating NEW HINT Buying Process (experience) -Sales people -Service -Facility (Eg: Internet) -Familys feeling Price sensitivity Finance issue
+ 26 Motivation STEP 1: Customers have more rights on review sheet New Review Sheet Personal Information Additional Questions Review Additional Rating
27 Additional Rating Description Matching Level
28 Additional Questions Are you satisfied with this purchase? Yes No Please tell us WHY (Mark all that apply) Quality of in-store service Quality of vehicle Price Financial Service (Insurance, cash deal, warranty cost, monthly payment) Location of dealer Sales People In-store facility In-store environment Can/Cannot get vehicle in quickly enough Service Department hours Alternative transportation not available (rental car, shuttle, etc.) Other______________
+ 29 Motivation STEP 2: More visual hint for customers Dealer Rating Best Seller Extended Warranty Label Dealer: Dealer:
+ 30 Motivation STEP 3: New Dealers Resume Lounge with snacks and Facilities: drink Free Wi-Fi Insurance Deal/Gift Services: Extended Warranty Maintenance Free certification More space for Dealer Description 3 others viewing this vehicle right now 15 others bought vehicles from this dealer in the last month
+ 31 Motivation STEP 4: Monthly rating list Monthly highest rating dealer Sub-rating list Monthly Star Dealer (label in the search page) Monthly recommended dealer Monthly best seller The most favorable dealer
+ 32 Better Cars.Com Combine Cars review sheet and Dealer review sheet Continue to review dealer? Yes No Dealers Review sheet Add schedule plug-in Online voicemail Add to Favorite Relevant vehicles to consider You Viewed Customer who view this also viewed Recommended base on your browsing history
+ Predictive Analytics for Cars.com Xiaofei Yang, Joyce Directed by Professor Martin Bariff Yi Shi, Ivy Yingqiu Zhu, Chris Thank Questions??? You for Your Listening

More Related Content

Predictive Analytics for Cars.com

  • 1. + Predictive Analytics for Cars.com Directed by Professor Martin Bariff Xiaofei Yang, Joyce Yi Shi, Ivy Yingqiu Zhu, Chris
  • 2. Introduction Data Preparation + Data Mining Methods Better Cars.com 2
  • 3. + Introduction Yingqiu Zhu, Chris 3
  • 4. + 4 INTRODUCTION SPSS MODELER
  • 5. + Data Preparation Yingqiu Zhu, Chris
  • 6. + 6 DATA PREPARATION KEY VARIABLES a. Basic Information b. Ratings c. Indicators
  • 7. + 7 DATA PREPARATION CLEANING Shopping Quality for Used of Repair Overall Rating Ready-to-Use Data
  • 8. + 8 DATA PREPARATION BALANCING Unbalanced Data Balanced Data 16.43% 30.21% 3.01% 5.54% 2.11% 3.88% 10.01% 18.40% 68.45% 41.96% 27 鐚
  • 9. + Data Mining Methods Yingqiu Zhu, Chris & Xiaofei Yang, Joyce
  • 10. + 10 METHODS K-Means Cluster Analysis Decision Tree Predictive Analysis Text Mining
  • 11. + 11 CLUSTER ANALYSIS K-MEANS K-Means is the most popular nonhierarchical clustering method. K-Means is based on geometric notions of similarity
  • 12. + 12 CLUSTER ANALYSIS K-MEANS Model Summary Model Accuracy
  • 13. R13 ANALYSI S Yes No 5 1 Importance 100% K-Means 5 1 5 1 + Yes No Importance 11%
  • 14. + 14 CLUSTER ANALYSIS K-MEANS Customers who purchase used cars from a dealer rating high for the dealer recommending the dealer to others.
  • 15. + 15 PREDICTIVE ANALYSIS INTRODUCTION OF DECISION TREE MODEL A tree-building algorithm Recursively splitting the data into smaller and smaller groups based on the fields that provides the maximum information gain Used C5.0 model in our research
  • 16. + 16 PREDICTIVE ANALYSIS DECISION TREE Decision Tree with Balanced Data 60% 40% Evaluation Standard Overall accuracy: > 80% Favorability between training & testing samples: The results with the testing sample compare favorably to the training sample.
  • 17. + 17 PREDICTIVE ANALYSIS DECISION TREE Rating - Customer Service
  • 18. + 18 PREDICTIVE ANALYSIS INTRODUCTION OF TEXT MINING MODEL Extract key concepts from the text and create categories with these concepts by using linguistic and frequency techniques Explore more information and context included in the data Better compare the results with and without text data and see the difference
  • 19. + 19 PREDICTIVE ANALYSIS DECISION TREE MODEL WITH TEXT DATA Model with Only Numeric Data Model with Text & Numeric Data
  • 20. + 20 PREDICTIVE ANALYSIS Interactive Workbench Categories and Concepts View
  • 21. 21 Trustful & Knowledgeable Extended warranty Relatively consistent pricing buying/buying process Nearest location < Service & Credibility
  • 22. + 22 PREDICTIVE ANALYSIS TEXT MINING Rating = 5 Rating = 1 High Overall Rating Low Overall Rating 100 90 100 80 90 80 70 70 60 60 50 50 40 40 30 30 20 20 10 10 0 0 % %
  • 23. + Better Cars.Com Yi Shi (Ivy)
  • 24. What we get from our analysis 24 Less influential dealers Purchase behavior information =>Higher overall rating Overall Rating Higher rating for Text Mining buying process & customer service =>New Hint for Consumer behavior =>Higher Overall Rating
  • 25. + 25 Stimulus Influence Overall Rating NEW HINT Buying Process (experience) -Sales people -Service -Facility (Eg: Internet) -Familys feeling Price sensitivity Finance issue
  • 26. + 26 Motivation STEP 1: Customers have more rights on review sheet New Review Sheet Personal Information Additional Questions Review Additional Rating
  • 27. 27 Additional Rating Description Matching Level
  • 28. 28 Additional Questions Are you satisfied with this purchase? Yes No Please tell us WHY (Mark all that apply) Quality of in-store service Quality of vehicle Price Financial Service (Insurance, cash deal, warranty cost, monthly payment) Location of dealer Sales People In-store facility In-store environment Can/Cannot get vehicle in quickly enough Service Department hours Alternative transportation not available (rental car, shuttle, etc.) Other______________
  • 29. + 29 Motivation STEP 2: More visual hint for customers Dealer Rating Best Seller Extended Warranty Label Dealer: Dealer:
  • 30. + 30 Motivation STEP 3: New Dealers Resume Lounge with snacks and Facilities: drink Free Wi-Fi Insurance Deal/Gift Services: Extended Warranty Maintenance Free certification More space for Dealer Description 3 others viewing this vehicle right now 15 others bought vehicles from this dealer in the last month
  • 31. + 31 Motivation STEP 4: Monthly rating list Monthly highest rating dealer Sub-rating list Monthly Star Dealer (label in the search page) Monthly recommended dealer Monthly best seller The most favorable dealer
  • 32. + 32 Better Cars.Com Combine Cars review sheet and Dealer review sheet Continue to review dealer? Yes No Dealers Review sheet Add schedule plug-in Online voicemail Add to Favorite Relevant vehicles to consider You Viewed Customer who view this also viewed Recommended base on your browsing history
  • 33. + Predictive Analytics for Cars.com Xiaofei Yang, Joyce Directed by Professor Martin Bariff Yi Shi, Ivy Yingqiu Zhu, Chris Thank Questions??? You for Your Listening

Editor's Notes

  1. Now we have general ideas about why people give dealers low or high rating? What do they really care about during purchase? Then how to help dealers improve that? Before talking about that. We should first stand in dealers&apos; place, thinking about what do dealer want? More customers, high ROI and better SRP. Dealers are always try to topped the list. Rather than just provide them lot of tips in dealers section, how about do something to motivate them. No one want to lose in competition. Step 1: give customers more rights on review sheet, not just simple three rating. Customers could evaluate more things about dealer
  2. In order to enable customers find more dealer information in most efficient way. Cars. Com could encourage dealer list more information on their resume. Like if they provide free WIFI or not. What kind of particular do they have, do they provide insurance service, extended warranty service, or maintenance service. Do they have any seasonal deal? Do they provide free certification service. Give dealer more space to descript themselves. Another way to encourage dealers is plug-in real time label in their page, tell customers how many people are viewing this vehicle right now and the dealers sales volume in the last month.
  3. The last efficient things to motivate dealers is post some rating list on the home page month by month. like.this could give customers more hint to search as well as encourage dealers to be better.