Using the same dataset leveraged for Portfolio Builder Exercise #1 & #2, write up a third report that answers the following:
- Apply a MARS model with all features.
- How does the model performance compare to your previous models?
- How many of the features are influential? Which 10 features are considered most influential?
- Does your model include hinge functions? If so, explain their coefficient and plot their impact on the predicted response variable.
- Does your model include interactions? If so, pick the interaction effect that is most influential and explain the coefficient.
- Apply a random forest model.
- First, apply a default random forest model.
- Now apply a a full cartesian grid search across various values of \(m_{try}\), tree complexity & sampling scheme.
- Now run a random grid search across the same hyperparameter grid but restrict the time or number of models to run to 50% of the models ran in the full cartesian.
- Pick the best performing model from above.
- Identify the most influential features for this model.
- Plot the top 10 most influential features.
- Do these features have positive or negative impacts on your response variable?
- Create partial dependence plots for these features. Explain the relationship between the feature and the predicted values.
🏠