After a few beers your CIO invited his buddy from Blue Moon consulting to propos

Question

After a few beers your CIO invited his buddy from Blue Moon consulting to propose a project using data mining to improve the targeting of the new service that you have been a principal in developing. The service has been quite successful so far, being marketed over the last 6 months via your ingenious, and very inexpensive, word-of-mouth campaign. You've already garnered a pretty large customer base without any targeting, and you've been seeing this success as your best stepping stone to bigger and better things in the firm. After some reflection, you've decided that your best course of action is to play a key role in ensuring the success of the data mining project as well. You agree with your CIO's statement in a meeting with Blue Moon, that very accurate targeting might cost-effectively expand your audience to consumers that word-of-mouth would not reach.

Identify the four most serious flaws in this abridged version of Blue Moon's proposal, and suggest how to ameliorate them. You can accept that Blue Moon has accurate information about the service.

"We will build a logistic regression (LR) model to predict service uptake for a consumer, based on the data on your existing customers, including their demographics and their usage of the service. We believe that logistic regression is the best choice of method because it is a tried and true statistical technique, and we can interpret the coefficients of the model to infer whether the attributes are statistically significant, and whether they make sense. If they are and they do, then we can have confidence that the model will be accurate in predicting service uptake. We will apply the model to our (Blue Moon's) large database of consumers, and select out those whom the LR model predicts to be the most likely to subscribe. It is a fixed-price, fixed-cost, fixed-term service, so this also will in effect rank them by expected profit as well."

Explanation / Answer

Ans: The most serious flaws are

1) Firstly, on target variable there isn't any past data, so the solution for this is to collect all this new data.

2) The model's accuracy need to be based on how good is prediciting for target varible. But not whether an attribute significant statistically.

3) The evaluation strategy is very unclear as we cant predict that the accuracy method is the best one. So we need to look precisly for this one.

4) Training and testing the data should be performed inorder to ensure that isn't overfit and well generalizes.

5) what percentage of the consumers would fall under "more likely to subscribe" should be established.

//// Thank You /////

Related Questions

Navigate

• Browse (All)
• Browse A
• Subjects

Integrity-first tutoring: explanations and feedback only — we do not complete graded work. Learn more.