MLA-C01 Question 124
Single answerA financial services company wants to build a machine learning model to predict whether a customer will default on a loan. The dataset provided contains categorical features such as 'Employment Type' and 'Loan Purpose', numerical features such as 'Annual Income' and 'Loan Amount', and a binary target variable 'Default' (1 for default, 0 for no default). The company needs a model that provides both high predictive performance and interpretability for regulatory compliance. Which modeling approach should you choose?
- A
Train a deep neural network with multiple hidden layers and ReLU activation functions
- B
Use a logistic regression model with one-hot encoding for categorical features
- C
Choose a random forest model with categorical features encoded as integers
- D
Use a gradient boosted trees model with target encoding for categorical features
Show answer and explanation
Correct answer: B
Explanation
The company's requirement for interpretability makes logistic regression the most suitable choice. Logistic regression inherently provides interpretable coefficients that can be easily explained to stakeholders and regulators. One-hot encoding ensures that categorical features are appropriately handled, making the model both interpretable and suitable for the binary classification task.
- A. Incorrect.
Deep neural networks are powerful for high-dimensional data and complex relationships but are not well-suited for interpretability, which is a key requirement in this scenario.
- B. Correct.
Logistic regression is a simple and interpretable model. Combined with one-hot encoding for categorical features, it meets the need for interpretability and performs adequately for binary classification tasks.
- C. Incorrect.
While random forests can handle categorical and numerical data well, they are not inherently interpretable, which is a critical requirement for regulatory compliance in this case.
- D. Incorrect.
Gradient boosted trees can achieve high predictive performance but are not easily interpretable without additional tools like SHAP or LIME. Thus, they do not meet the interpretability requirement.