Question 1.1. A supervised machine learning model for predictive analytics requires that we specify a data set of observations from which to learn, a set of input variables that describe each observation, and an outcome variable. In addition, we need labeled observations from which to learn a relationship between the input variables of each observation and the corresponding outcome variable. In this Apple setting, what are input variables that describe an observation? Select all that apply (you may select none of the answers if applicable): A picture uploaded to Apple iCloud (the digital image) an Apple iCloud account holder id A picture uploaded to the NCMEC database (the digital image) is this picture illegal (yeso) Is this picture child pornography (yeso) "neural hash" of a user uploaded picture "neural hash" of a NCMEC picture Question 1.2. Apple's SVP of software engineering notes that: "This isn't doing some analysis for, did you have a picture of your child in the bathtub? Or, for that matter, did you have a picture of some pornography of any other sort? This is literally only matching on the exact fingerprints of specific known child pornographic images." Mr. Federighi asserts that this system is not a predictive classifier using supervised machine learning. Select the one best answer: This is not predictive analytics because it measures similarity and we use similarity in descriptive analytics. This is supervised machine learning because it uses the NCMEqdatabase as training data to classify what is and is not child pornography. This is NOT supervised machine learning because while there is a source of training data, there is no source of validation and test data. T is predictive analytics because it uses past knowledge of child pornography to predict whether newly photographed behavior qualifies as child pornography. None of the above 3.4 The consulting company asks you to improve the accuracy of the survey-based classification. What might you try (select all that apply). Redesign the survey and include additional questions Redesign the survey and remove questions Change the cutoff that translates the survey score into a positive or negative prediction Change the ratio of labeled data when partitioning to increase the amount of test data relative to training and validation None of these Problem 1. In 2021, Apple announced a new feature to identify child pornography in photos stored using Apple's iCloud Photos. We might think of this as a basic tool for predictive classification: When a user uploads an image to iCloud, that image is either classified as "child pornography" or "not child pornography." Joanna Stern, a columnist for the WSJ, interviewed Craig Federighi, Apple's senior vice president of software engineering and wrote about the feature in the April 13, 2021 edition: How does this work? Some basics: The National Center for Missing and Exploited Children (aka NCMEC) maintains a database of known illegal child pornography. Other big tech companiesGoogle, Facebook, Microsoft, etc. - have methods of scanning photos you upload to their servers to see if any match the images in the NCMEC repository.... The il. al photos collected by NCMEC and other similar child-safety organizations have been converted into cryptographic codes-strings of numbers called "neural hashes"-that identify signature chaacteristics of images. After this update hits your phone, in an IOS update due sometime this year, the software will generate hashes for your own photos as they're prepared for upload to iCloud. The device then would cross reference your image hashes to the hashes from the child-pornography database.... The system is designed to it atch only fingerprints of known illegal images.... Even if there is a positive match, no alarm bells ring at this point. However, if an account collects around 30 safety vouchers corresponding to illegal images, according to Mr. Federighi, the account gets flagged to Apple's human moderators. They review the vouchers (and no other images) to see if they actually contain potentially illegal images. If they de Apple reports the account to NCMEC. Problem 3. Employee Churn Every year, a major consulting company conducts a satisfaction survey of their employees. For each employee, the survey results in a probability predicting whether that employee will Stay (True) or Go (False). High employee satisfaction (Positive prediction) predicts who will stay. Low employee satisfaction (Negative prediction) predicts that people will leave. Consider the following confusion matrix. Each column represents survey responses. Each row represents those employees who actually remained (Stay) or left (Go). Consider the comparison of two different satisfaction surveys. Suppose that every employee completes both Survey 1 and 2