AIF-C01 Question 184
Select 3A healthcare company is using a machine learning model to predict the likelihood of patients developing a specific medical condition. The company is required to comply with industry regulations demanding that medical decisions based on AI outputs are explainable to both patients and practitioners. Which of the following approaches would help the company ensure their model is transparent and explainable?
- A
Use a simpler, interpretable algorithm such as linear regression instead of a complex deep learning model.
- B
Incorporate SHAP (SHapley Additive exPlanations) values to provide feature importance for predictions.
- C
Use an ensemble method like Random Forest without additional explanation tools, as it provides high accuracy.
- D
Provide detailed documentation of the model's training data and feature selection process.
- E
Implement LIME (Local Interpretable Model-agnostic Explanations) to generate explanations for individual predictions.
Show answer and explanation
Correct answers: A, B, E
Explanation
Transparent and explainable models are crucial in regulated industries like healthcare to ensure that AI decisions can be understood and trusted by practitioners and end users. Approaches like using inherently interpretable models, incorporating explanation tools such as SHAP and LIME, and reducing model complexity promote transparency. While documentation is beneficial, it does not directly contribute to explaining model predictions to stakeholders.
- A. Correct.
Simpler, interpretable algorithms like linear regression are easier to understand and explain to stakeholders compared to complex models. This makes them suitable for use in scenarios where transparency is a priority.
- B. Correct.
SHAP values are a well-known method to provide clear insights into the contribution of each feature to a model's predictions, enhancing the explainability of the model.
- C. Incorrect.
While Random Forest often offers high accuracy, it is not inherently transparent or explainable without additional tools, making it unsuitable on its own for this use case.
- D. Incorrect.
Providing documentation of training data and feature selection is helpful for reproducibility but does not directly enhance the explainability of model predictions to non-technical audiences.
- E. Correct.
LIME is a popular tool for generating local explanations for individual predictions, which can be used to provide transparency and interpretability for complex models.