Exam AWS Certified Machine Learning - Specialty topic 1 question 1 discussion

The Answer is A. Reasons: 1. accurate is 86% 2. FN=4, FP= 10. The question is asking why this is a feasible model which means why this is working. So it is not asking the explaination of the unit cost of churn(FN) is greater than cost of incentive(FP). It is asking from the matrixs result, the number it self, FN(4) is less than FP(10). The model successfully keep a smaller number of FN regarding of FP.

Such question cannot be answered because we do not know how much more is greater the cost of churn than the cost of the incentive. CoC - Cost of Churn CoI - Cost of Incentive cost incurred by the company as a result of false positives = CoI * 10 cost incurred by the company as a result of false negatives = CoC * 4 So is it the case that CoI * 10 > CoC * 4 => CoI > 0.4 * CoC, or rather CoI < 0.4 * CoC? We don't know that because we don't know what does it mean "far greater", is it 100% greater, or is it 500% greater or any other number.

Option C is indeed the correct choice. The model is 86% accurate, and the cost of false positives (offering incentives) is less than the cost of false negatives (losing customers). This makes the model viable for production.

Changing my earlier Answer from A to C. Cost of FP(10) is lower than Cost of FN(4)

some people that voted A have the right idea, but they chose the wrong option because they need to read the question again. We all agree that the cost of churn is much higher. So a false-negative means a customer churned and you didn't do anything about it (because your model said "churn=no") . A false positive means you tried to keep a customer that was not going to leave anyway (because your model said "churn=yes"). As you can see, false-negative is way costlier and should be avoided, therefore answer is C.

Cost incurred for churn higher than incentive. Cost of FN is higher than FP. And accuracy is 86%.

what tomatoteacher said

Accuracy is 86% and it should be A or C. Lost is very high compare to intensive. Means it is Okay to give intensive to customers who are not going to leave. Which means False positives potion.

Should be A. Since the cost of churn is much higher, the priority should be focused on minimizing FN and a viable model should be one with FN < FP, isn't it?

Definitely C. If you look at the same question in https://aws.amazon.com/blogs/machine-learning/predicting-customer-churn-with-amazon-machine-learning/. Same question, but the confusion matrix is flipped in this case( TP top left, Tn bottom right) . When you miss an actual churn (FN) this would cost the company more. Therefore the answer is C 100%. I will die on this hill. I spent 20 minutes researching this to be certain. Most people who put A are incorrectly saying FPs are actual churns that are stated as no churn.. that is what a FN is. You can trust me on this.

There are more FP's than FN's, however the costs of FN's are far larger than that of FP's. So: numberof(FP) > numberof(FN), costperunit(FP) << costperunit(FN). This itself could suggest that totalcosts(FP) < totalcosts(FN), but would be somewhat subjective, since it is not stated how far the unitary costs are. What is suggested, however, is that the model is indeed viable (question asks WHY the model is viable, and not WHETHER it's viable). If the model didn't exist, there would be no way that there are FP's or FN's, but churns would still exist, which have the same cost as FN's. So it means the total costs with FP's must be less than the total costs with FN's (churns).

Correct Answer C. Explanation: The model's accuracy is calculated as (True Positives + True Negatives) / Total predictions, which is (10 + 76) / 100 = 0.86, or 86%. The cost of false positives (customers predicted to churn but don't) is less than the cost of false negatives (customers who churn but were not predicted to). Offering incentives to the false positives incurs less cost than losing customers due to false negatives. Therefore, this model is viable for production.

A. NO - accuracy is TP+TN / Total = (76+10)/100 = 86%; we know the model is working, so the cost of giving incentives to the wrong customers (FP) is less than the cost of customers we missed (FN), cost(FP) < cost(FN) B. NO - accuracy is 86%, precision is TP / (TP+FP) = 10 /(10+10) = 50% C. YES - accuracy is TP+TN / Total = (76+10)/100 = 86%; we know the model is working, so the cost of giving incentives to the wrong customers (FP) is less than the cost of customers we missed (FN), cost(FP) < cost(FN) D. NO - accuracy is 86%, precision is TP / (TP+FP) = 10 /(10+10) = 50%

C is the correct answer

FN has a higher cost than FP, so A is a better choice than C.

A) Because FN = 4 < FP = 10. FN are missed churns, and FP is misidentified churns.

A is the correct answer