Blog

Assume you test the decision tree created above on a sample of 100 observations and obtain the following confusion matrix. In the testing data, you have 80 NO SPAM and 20 YES SPAM.       ACTUAL CLASS     YES SPAM NO SPAM PREDICTED CLASS YES SPAM 5 (TP) 5 (FP) NO SPAM 15 (FN) 75 (TN)

Which of the following statements about the Entropy of the partition/node “Do you know the sender?” = NO (the green node with YES SPAM, 0.61 and 6%) is TRUE?

Which of the following statements about Decision Trees is FALSE? 

Which of the following statements is FALSE about the difference between unsupervised and supervised learning?

Which of the following statements about lift in Association Rules is FALSE?

Finally, assume you test the new decision tree (let us call it, DT2) on the same testing data as the first decision tree (let us call it, DT1). You obtain this new confusion matrix:     ACTUAL CLASS     YES SPAM NO SPAM PREDICTED CLASS YES SPAM 10 (TP) 25 (FP) NO SPAM 10 (FN) 55 (TN)   The accuracy for DT2 is 65%. When you report this number to your team, they are worried the model is not good enough and they suggest going back to DT1 instead. You need to convince them they are not thinking about it in the right way and that, instead, DT2 could be more suited in this context. To make your argument convincing, you should present them with additional performance measures and explanations about the models’ performance and how to use/interpret such performance measures. You need to choose which performance measures to use in your argument – keep in mind the business scenario presented at the beginning!    

Your colleagues notice that the tree is classifying only a total of 10 emails as YES SPAM. They suggest revisiting the way in which the decision tree makes the classifications; in particular, they would like to increase the number of emails classified as YES SPAM. Which of the following actions will allow you to do so?

Assume your company would like to group customers into segments based on their purchasing behavior. Along with a colleague, you decide to run k-Means to create the segments. You and your colleague are discussing how many clusters to use, and you decide to create the Elbow plot (using WSS only) to help with the decision. You present the plot to your colleague. Your colleague suggests using k = 8 as final number of clusters, because that is the point where the WSS is at its lowest. Do you agree with your colleague? Why or why not? In your answer, make sure to clearly explain what the WSS captures and how to use the elbow plot to figure out a “good” number of clusters.    

Which method/algorithm, among the ones listed below, is the most appropriate to address this data-mining task? Create groups of stocks based on data about their fluctuations. You know that your data has a spherical shape.

Q4