IST718 WK3
3.1 Scenario Introduction
- How does a customer make a choice about phones
- How does a customer make a choice about groceries
NOTES: Walmart removes bananas from market basket analysis
- How do we retain customers
- More choices
- More value (lower prices)
3.2 Data Review
3.3 Model Review
3.4 Recommendation
NOTES: So as we start to look at potential relationships and potential marketing tie-ins, we’ve got some marketing tie-ins between dairy products and vegetables. We’ve got marketing tie-ins between dairy products and the fresh baked bread. If we look at it from the graph perspective, we get another way to view this. The graph also gives us insight into how beef and dairy products can be related and how the bread and the vegetables can be related. From both of these graphs, we see this high association for market baskets and containing the fresh baked bread items.
3.5 Code Review
We’re going to build a model that allows us to determine the likelihood of a customer defecting or transitioning from one plan to another.
LOGIT MODEL
And we can see running through the logit regression, the pseudo R-squared– and it really is a pseudo R-squared for logistic regression. You probably want to use AIC or another statistical measure. One thing you can look at, especially the way it’s charted out in Python, is if any of these coefficient values cross 0, then that coefficient is probably not statistically significant in that data set.
ASK YOURSELF:
- Why is a relationship here?
- Why might a relationship be missing?
TODO:
Try to recreate the graphs from Chapters 3 & 4 in Modeling Techniques in Predictive Analytics
3.6 Customer Choice/Preference
- Retaining
- Switching
–
LOGISTIC REGRESSION
Consumer choice– one of the things we can look at is, for the customer, what are the types of variables that would cause a customer to switch. As we start to look at the preference in a 0 or 1 decision, we switch from using linear regression to a logistic regression, in order to be able to model this decision of yes or no.
3.7 Product Promotion
Large Sparse Matrix
A little bit of terminology. We’ve got an item set. The item set is the items that are purchased together. We’ve got an association rule, which is the antecedent and the consequent. So in this case, the fresh bread would be the antecedent, the thing that’s on the left-hand side. And then, the consequent would be the item that is purchased with the fresh bread, whether it be vegetables, whether it be dairy, on the right-hand side. This does get into a little case of the chicken and the egg, which one is actually correlated with the other, and it really becomes a matter of frequency and priority when looking across not just one individual market basket but all the market baskets that are in the transaction set.
Selection criteria. We can talk a little bit about support, lift, and confidence definitely in your reading. But as we can adjust the statistical measures of support, confidence, and lift, we can either increase the amount of rules we get or decrease the amount of rules we get. And when we say rules, it’s these association rules. If I buy bread, I also buy milk.