Exam AWS Certified Machine Learning - Specialty topic 1 question 72 discussion

AC - correct answer

I think ACE are all correct

A. YES - standard for overfitting B. NO - we have already too much overfitting C. YES - feature elimination can reduce model complexity and thus overfitting D. NO - that does dimensionnality reduction to 2D or 3D, for visualization; we want more than a few features E. NO - LDA is an alternative to logistic regression; it may not address overfitting

A due to fitting C Recursive feature elimination (RFE) is a wrapper method that iteratively removes features based on their importance scores from a classifier. RFE starts with all features and then eliminates the least important ones until a desired number of features is reached. This can help to reduce the dimensionality of the dataset and improve the model performance by removing irrelevant or redundant features. The Marketing team can then interpret the model by looking at the remaining features and their importance scores.

AC are the correct

How can we add features to the dataset provided.... we can't make them up from thin air. Hopefully the moderators can provide some insight on this. I was thinking of paying for this site but the answers are all over the place.

A. Add L1 regularization to the classifier and C. Perform recursive feature elimination are the methods that can be used to improve the model performance and satisfy the Marketing team's needs. Explanation: A. Adding L1 regularization to the logistic regression classifier can help to improve the model performance and reduce overfitting. This can also help to highlight the relevant features for churn prediction as L1 regularization can shrink the coefficients of irrelevant features to zero. C. Recursive feature elimination can be used to select the most relevant features for the model. This can help to improve the model performance and highlight the relevant features for churn prediction.

A. Adding L1 regularization can help to reduce overfitting by shrinking the coefficients of less important features towards zero, which can improve the model's generalization performance on the validation set. C. Recursive feature elimination is a feature selection technique that removes the least important feature at each iteration and trains the model on the remaining features until a desired number of features is reached. This method can be used to identify the most relevant features for the prediction task and reduce the dimensionality of the dataset, leading to improved model performance and interpretability for the Marketing team.

AC - Key: logistic regression model = non linear in terms of Odds and Probability, however it is linear in terms of Log Odds. Key: Large gap between training & validation = overfitting => 5 techniques to prevent overfitting: 1. Simplifying the model | 2. Early stopping 3. Use data argumentation | 4. Use regularization | 5. Use dropouts A - yes to avoid overfitting (although i am thinking it is talking about regressor) Not B - add feature will lead to overfitting C - feature elimination - prevent overfitting Not D - t-SNE is a nonlinear dimensionality reduction technique Not E - find feature correlation only - Linear discriminant analysis (LDA), normal discriminant analysis (NDA), or discriminant function analysis is a generalization of Fisher's linear discriminant, a method used in statistics and other fields, to find a linear combination of features that characterizes or separates two or more classes of objects or events.

L1 won't do naturally the feature elimination? I guess AB

why not A & D? or C & D? does not t-SNE grant the marketing team's wish for visualization of relationships? or are we to presume that A&C are best as C (recursive feature elimination) grants us some visualization of feature importance.

AC is correct

overfitting: add regularization, remove features