Analytics and Machine Learning Case Solution
- [5 points] A bank has hired two external competing consulting firms to help it predict the balance that is carried over by customers who open a new credit card account with them by switching from other banks. The company gives its large database (the training data) of past customers who had switched to them, with the accompanying balance carried over and with related X variables/features to the two competing consulting firms and each of the firms builds a prediction model for the balance carried-over.
Consulting firm #1 reports that based on the training data, the training mean squared error of prediction of the irmodelis 305,while consulting-firm#2 reports that based on the raining data, the training mean squared error of prediction of their model is 160.
The head of the Business Development group at the bank is impressed that the model of consulting firm #2 has a training mean square error that is almost half that of the model of firm #1 and recommends that the company retain consulting firm #2.
Would you support this recommendation?(You MUST provide a BRIEF yet CLEAR just unification for your choice below. 4 points for the justification)
- Yes (ii) No
Firm #2 is chosen as it has more accurate model as compared to firm #1. The mean squared error of firm #2 is less than that of firm #1. The difference between actual values and predicted are very close.
- 5 points] A data scientist is interested in predicting the average annual balance that credit cardholdershaveontheircards.Hehasaccessto400featuresoneachcustomerthathewishes touseinhisregressionmodeltopredictthebalance.Sincehehasalargenumberoffeatures,he uses the LASSO method with the𝜆𝜆(lambda)value-selected by conversationalists up with a regression model that has only 52 features in it out of the 400 that he started with.
Suppose you trained a regression model on the same data using the LASSO, but used a 𝜆𝜆 value that was substantially larger than the one used for his model, would you end up with a model that had
- More features than the 52 in his model?
- Fewer features than the 52 in his model?
(you must justify your answer briefly to get full credit)
There will be more features than the 52 in this model, the reason behind is that if lambda decreases in value from one(1) to zero(0) for each iteration then it will select more columns by cross-validation.
- [5 points] Suppose you are interested in converting the following 3 text documents into numerical values so that you can incorporate the data into a model and so compute aTF-IDF score for each word and document-combination
Doc 1: The pharma sector has done well in the last year
Doc 2: Analysts have downgraded the pharma sector for the next year
Doc 3: The tech sector has lagged behind the pharma sector; however, tech has outperformed manufacturing by a large margin
- Will the TF-IDF score of “pharma” be high across all documents?
No since it is present in all three documents with equal occurrence then it will yield an IDF of 0 which when multiplied by TF of each document will result in 0.
- Will the TF-IDF score of “tech” be high in document 3?
Yes as it yields a TF of 1/9 and and IDF of log(3).
- Will the TF-IDF score of “tech” be high in document 1?
No as the TF is equivalent to 0.
For-each question above,provide“yes”or“no”answer with as centerboards justification. Zero credit will be given if there is no justification provided
An analytics start-up has generated a data base of costumes,living in NY and NJ,whose pat a grocery store chain and who possess a loyalty card from the chain. Since they can identify the customer through the loyalty card, the start-up has for each customer not only the average monthly$amount-spent by the customer at the store but,by accessing and merging in formation from various databases,also aloe to featureless the customer(demographics,salary,area where they live,number of times they visit the store,the time of day when they visit the store,etc).The start-up wishes to use this training data set to build a predictive model to predict the average monthly $ amount spent at the store by a loyalty card owning-customer.
- [1 point] Is this is a supervised or unsupervised learning problem? If it is a supervised learning problem, what is the Y variable?
This is supervised learning problem. The loyalty status of customer is treated as the Y variable.
- [4 points] As mentioned above, the training data is of loyalty card customers living in NY and NJ. The grocery chain has recently expanded into the states of Iowa, Oklahoma, Illinois and Kansas and is slowly trying to build up and maintain its client base. In order to do so, it plans to identifytheloyaltycardcustomersinthesestateswhoarepredictedtospendthelargestaverage monthly $ amount at their stores and then mail them discount coupons. Towards this end, the chain uses the prediction model built by the start up from the training data mentioned-earlier.
In a few sentences, briefly, explain with justification whether you think this is a reasonable idea or not
Firstly this improves customer relation and loyalty towards the store. It enhances the shopping experience as customer can purchase more items within the allotted amount per customer. Customer will become more frequent and regular. Ultimately this will attract more customers due to word of mouth and advertisements to spend more in order to obtain discount coupons. Sometimes, certain products don’t work well: this could be for a number of reasons, ranging from price to product visibility. By offering coupons, redeemable on such products, you will be re-introducing them to consumers. Coupons in one way can mitigate the advertisement cost.
- [10 points] You are running a company that specializes in selling high end cigars. To help you catalog and describe your collection of cigars when making suggestions to customers, you hired cigar specialist who explains that some of the key features to look for in a cigar are its sheen, moisture and flavor intensity. She tells you that she can rate every cigar along each of these 3 dimensions on a scale of1-5.
A regular customer, who has been buying a lot of cigars of one kind from you, wishes to try out another cigar which is similar to the one he has been buying. The cigar that he has been buying, according to the specialist’s ratings, has a sheen rating of 3, a moisture rating of 4 and a flavor intensity rating of 2.
Amongst the cigars that you stock ,you have 2 cigars who seating for the 3 factor arrears follows: Cigar 1: sheen rating = 4, moisture rating = 2, flavor intensity rating =4
Cigar 2: sheen rating =5, moisture rating=3, flavor intensity rating=3
Which of the 2 cigars above would you recommend to the regular customer? You must justify your choice briefly yet clearly. Zero credit if no justification is provided.
Customer Preference: 3,4,2
Cigar1:4,2,4 => 4-3,|2-4|,|4-2|=>1,2,2
Cigar2:5,3,3 => 5-3,|3-4|,|3-2|=>2,1,1
From the above calculations we can see that cigar2 has a closer match to that of customer preference. Cigar2 will be recommended on this basis so as not to dissatisfy the customer…………………..
This is just a sample partial case solution. Please place the order on the website to order your own originally done case solution.