Please do not include AI , Handwritten is prefreable Question 1 You run the fraud team at a major bank. You have implemented a new system that uses artificial intelligence (AI) to identify transactions that might result from a scam. However, the AI is not perfect: The AI correctly flags 80% of scam transactions as scams (the true positive rate, or sensitivity). The AI correctly flags 99% of legitimate transactions as not being scams (the true negative rate, or specificity). You also know that 1 in 10,000 transactions relate to a scam. a) The bank seeks to decide whether it should automatically block transactions flagged as a scam by the AI. To inform their decision, they would like to know the probability that a transaction flagged as a scam by the AI is actually a scam. Calculate the probability using Bayes' rule. b) On hearing your answer, the head of the retail bank wants to better understand the intuition behind it. Use natural frequencies to show the probability that a transaction flagged as a scam by the AI is actually a scam. c) The bank wants to adjust the AI system to improve its performance. If they could only improve one of the following, which would have a greater effect on the probability that a flagged transaction is actually a scam: increasing the true positive rate (sensitivity) from 80% to 99%, or increasing the true negative rate (specificity) from 99% to 99.9%? You do not need to make any mathematical calculations in answering this question (although you are welcome to if it helps.) Explain the intuition.

Please do not include AI , Handwritten is prefreable

Question 1

You run the fraud team at a major bank. You have implemented a new system that uses artificial intelligence (AI) to identify transactions that might result from a scam. However, the AI is not perfect:

• The AI correctly flags 80% of scam transactions as scams (the true positive rate, or sensitivity).
• The AI correctly flags 99% of legitimate transactions as not being scams (the true negative rate, or specificity).

You also know that 1 in 10,000 transactions relate to a scam.

a) The bank seeks to decide whether it should automatically block transactions flagged as a scam by the AI. To inform their decision, they would like to know the probability that a transaction flagged as a scam by the AI is actually a scam. Calculate the probability using Bayes' rule.

b) On hearing your answer, the head of the retail bank wants to better understand the intuition behind it. Use natural frequencies to show the probability that a transaction flagged as a scam by the AI is actually a scam.

c) The bank wants to adjust the AI system to improve its performance. If they could only improve one of the following, which would have a greater effect on the probability that a flagged transaction is actually a scam: increasing the true positive rate (sensitivity) from 80% to 99%, or increasing the true negative rate (specificity) from 99% to 99.9%? You do not need to make any mathematical calculations in answering this question (although you are welcome to if it helps.) Explain the intuition.