You work for a major banking and financial services firm as a machine learning specialist.
Your firm has decided to improve its fraud detection for specialized cases where fraudulent actors attempt to open accounts through your firm's banking and trading services.
These services have websites where potential customers can open accounts by completing online forms.
These services make use of your firm's highly secure customer and account data stores. You have been assigned the task of determining when a known fraudulent actor attempts to open a new account.
You have decided to build a machine learning solution to solve this problem.
Since your firm has a very large customer base, several million customer accounts, you need to consider the performance and the precision of your fraud detection process. You have decided to use the AWS Glue FindMatches ML Transform to process your online form data to find matching known fraudulent accounts in your firm's data stores.
Knowing that detecting a fraudulent actor is of primary importance, how should you configure the AWS Glue FindMatches ML Transform parameters to achieve the most performant and accurate fraud detection process?
Click on the arrows to vote for the correct answer
A. B. C. D.Answer: C.
Option A is incorrect.
Setting the FindMatches precision-recall parameter to ‘precision' minimizes false positives (when you don't have a match of a fraudulent account but mark it as a match mistakenly)
But you are more concerned about minimizing false negatives (when you have a match of a fraudulent account but fail to detect it).
Option B is incorrect.
Setting the FindMatches precision-recall parameter to ‘precision' minimizes false positives (when you don't have a match of a fraudulent account but mark it as a match mistakenly)
But you are more concerned about minimizing false negatives (when you have a match of a fraudulent account but fail to detect it).
Option C is correct.
Setting the FindMatches precision-recall parameter to ‘recall' minimizes false negatives (when you have a match of a fraudulent account but fail to detect it)
This is what you want.
Also, setting the FindMatches accuracy-cost parameter to ‘accuracy' maximizes the transform accuracy of finding matching records as fraudulent.
Option D is incorrect.
Setting the FindMatches precision-recall parameter to ‘recall' minimizes false negatives (when you have a match of a fraudulent account but fail to detect it)
This is what you want.
But, setting the accuracy-cost parameter to ‘lower cost' favors cost or the speed of running the transform at the expense of the transform's accuracy.
This may make your transform more performant, but your primary concern is detecting a fraudulent actor.
So you should set the accuracy-cost parameter to ‘accuracy'.
Reference:
Please see the AWS Glue developer guide titled Machine Learning Transforms in AWS Glue, and the AWS Glue developer guide titled Tuning Machine Learning Transforms in AWS Glue.
In order to configure the AWS Glue FindMatches ML Transform parameters to achieve the most performant and accurate fraud detection process, we need to consider the precision-recall trade-off and the accuracy-cost trade-off.
Precision-Recall trade-off: Precision is the ratio of true positive cases to the total positive cases predicted by the model, while recall is the ratio of true positive cases to the total actual positive cases. In fraud detection, it is important to detect all actual fraudulent cases (high recall) but also to minimize false positives (high precision).
Accuracy-Cost trade-off: Accuracy is the ratio of correct predictions to the total number of predictions, while cost refers to the resources (such as time, compute power, or storage) required to make those predictions. In fraud detection, it is important to minimize false positives (to avoid unnecessary investigation costs) while also minimizing false negatives (to avoid missed fraudulent cases).
With these considerations in mind, the best configuration for the AWS Glue FindMatches ML Transform parameters would be to set the precision-recall parameter to ‘precision' and the accuracy-cost parameter to ‘lower cost' (Option B).
By setting the precision-recall parameter to ‘precision', we prioritize minimizing false positives and maximizing precision in order to reduce unnecessary investigation costs. This is because we want to avoid flagging too many legitimate customers as potentially fraudulent.
By setting the accuracy-cost parameter to ‘lower cost', we prioritize minimizing the resources required to make predictions. This is because we have a very large customer base and need to ensure that our fraud detection process is scalable and efficient.
Option A, setting the accuracy-cost parameter to ‘accuracy', would prioritize accuracy over cost, which may lead to higher false positives and unnecessary investigation costs.
Option C, setting the precision-recall parameter to ‘recall', would prioritize detecting all actual fraudulent cases over minimizing false positives, which could lead to flagging too many legitimate customers as potentially fraudulent.
Option D, setting both parameters to ‘lower cost', would prioritize minimizing resources over both precision and recall, which could lead to missed fraudulent cases or too many false positives.