Investigating Suspicious Transactions with Mimic Explainer
Question
You are working for a company which is operating a webshop.
All the transactions flowing through the site are directed to a real-time inferencing web service to identify potentially risky transactions.
One of the transactions is classified by the model as “suspicious” and, before taking actions, you are tasked to investigate which features made the model “think” so.
You decide to use Mimic Explainer to help you understand why this specific transaction has been classified as “suspicious”
Does it serve your purpose?
Answers
Explanations
Click on the arrows to vote for the correct answer
A. B.Answer: A.
Option A is CORRECT becauseAzure offers a selection of model explainers: Tabular, Mimic and Permutation Feature Importance.
All of them can be used for explaining global importance of features, but only two of them (Tabular, Mimic) are applicable if you need to interpret local importance.
Mimic is a good choice for your task.
Option B is incorrect because either Mimic or Tabular explainer can be used for interpreting the local importance of features, i.e.
Mimic is a good choice.
Reference:
Answer: A. Yes
Explanation:
Mimic Explainer is a tool that helps data scientists understand how a machine learning model makes predictions by generating local and global feature importance scores. In this case, the data scientist wants to understand why a specific transaction has been classified as "suspicious" by the model. Mimic Explainer can help by generating feature importance scores, which can indicate which features had the most impact on the model's decision.
By using Mimic Explainer, the data scientist can get a better understanding of the specific features that contributed to the transaction being classified as "suspicious." This information can then be used to further investigate the transaction and potentially take actions to address any risks that may be associated with it.
Therefore, the use of Mimic Explainer can serve the purpose of understanding why the model classified the transaction as "suspicious."
