Homework 4

**Due Friday, April 14th at 11:59pm CST on Moodle.

Deliverables: Submit a single PDF containing your responses for Course Engagement.

Project Work

There is no new project work for this assignment. However – you should begin thinking about some classification models you can use for your Final Project.


Ethics in ML

Read the article How to Support your Data Interpretations. Write a short (roughly 250 words), thoughtful response about the ideas that the article brings forth. Which pillar(s) do you think is/are hardest to do well for groups that rely on data analytics, and why?

Portfolio Work


Revisions: Make any revisions desired to previous concepts. Make it clear to the preceptors what you want re-read (highlighted, commented, etc.)


New concepts to address:

Decision trees:

  • Algorithmic understanding:

    • Consider a dataset with two predictors: x1 is categorical with levels A, B, or C. x2 is quantitative with integer values from 1 to 100. How many different splits must be considered when recursive binary splitting attempts to make a split? Explain. (2 sentences max.)
    • Explain the “recursive”, “binary”, and “splitting” parts of the recursive binary splitting algorithm. Make sure to discuss the Gini index as a measure of node (im)purity and how it is defined differently for classification and regression trees.

  • Bias-variance tradeoff: What tuning parameters control the performance of the method? How do low/high values of the tuning parameters relate to bias and variance of the learned model? (3 sentences max.)

  • Parametric / nonparametric: Where (roughly) does this method fall on the parametric-nonparametric spectrum, and why? (3 sentences max.)

  • Scaling of variables: Does the scale on which variables are measured matter for the performance of this algorithm? Why or why not? If scale does matter, how should this be addressed when using this method? (3 sentences max.)

  • Computational time: Recursive binary splitting does not find the overall optimal sequence of splits for a tree. What type of behavior is this? What method have we seen before that also exhibits this type of behavior? Briefly explain the parallels between these methods and what implications this have for computational time. (5 sentences max.)

  • Interpretation of output: Explain the rationale behind the variable importance measures that decision trees provide. (4 sentences max.)

Bagging & Random Forests:

  • Algorithmic understanding: Explain the rationale for extending single decision trees to bagging models and then to random forest models. What specific improvements to predictive performance are being sought? (5 sentences max.)

  • Bias-variance tradeoff: What tuning parameters control the performance of the method? How do low/high values of the tuning parameters relate to bias and variance of the learned model? (3 sentences max.)

  • Parametric / nonparametric: Where (roughly) does this method fall on the parametric-nonparametric spectrum, and why? (3 sentences max.)

  • Scaling of variables: Does the scale on which variables are measured matter for the performance of this algorithm? Why or why not? If scale does matter, how should this be addressed when using this method? (3 sentences max.)

  • Computational time: Explain why cross-validation is computationally intensive for many-tree algorithms. What method do we have to reduce this computational burden, and why is it faster? (5 sentences max.)

  • Interpretation of output: Explain the rationale behind the variable importance measures that random forest models provide. (4 sentences max.)