A company is comparing a high-accuracy complex model with a simple, easy-to-explain model when selecting a credit model. Which is the MOST appropriate description of the general relationship between a model's performance (accuracy) and interpretability (ease of explanation)?
Choosing the general relationship between performance and interpretability.
In general, the more high-performing and complex the model, the harder it is to interpret; there is a trade-off.
Correct. When a model is made more complex to pursue higher performance, there is generally a trade-off in which its internal reasoning becomes harder to follow and interpretability decreases.
The higher-performing the model, the easier it is to interpret.
Generally it is the opposite: the higher-performing the model, the more complex its structure and the harder it is to interpret.
Performance and ease of interpretation do not rise together, so it is incorrect.
Performance and interpretability are independent and do not affect each other.
There is a trade-off relationship in which making a model complex to raise performance makes interpretation harder.
They are not unrelated to each other, so it is incorrect.
The more you raise interpretability, the more performance also improves.
Simple, easy-to-interpret models (such as linear regression) struggle to capture complex patterns and tend to be disadvantaged in performance.
Interpretability and performance do not rise together, so it is incorrect.
Remember the correct answer, the 'trade-off between performance and interpretability.'
・In general, the more a model is made complex to pursue high performance (for example, deep neural networks), the harder the internal reasoning is to follow and the lower the interpretability.
・For uses where accountability matters, there is also the judgment to choose an easier-to-interpret model even at some cost to performance.
'Higher performance, easier,' 'completely unrelated,' and 'raise interpretability and maximize performance too' are all errors that contradict the general relationship.