Building better models

• So far, in the labs, we’ve learned how to make predictions using the line of best fit, also known as linear models or regression models.
• We’ve also learned how to measure our model’s prediction accuracy by cross-validation.
• In this lab, we’ll investigate the following question:

Will including more variables in our model improve its predictions?

Divide & Conquer

• Start by loading the movie data and split it into two sets (See Lab 4C for help).
  • A set named training that includes 75% of the data.
  • A set named testing that includes the remaining 25%.
  • Remember to use set.seed.
• Create a linear model, using the training data, that predicts gross using runtime.
  • Compute the MSE of the model by making predictions for the testing data.
• Do you think that a movie’s runtime is the only factor that goes into how much a movie will make? What else might affect a movie’s gross?

Including more info

• Data scientists often find that including more relevant information in their models leads to better predictions.
  • Fill in the blanks below to predict gross using runtime and reviews_num.

lm(____ ~ ____ + ____, data = training)

• Does this new model make more or less accurate predictions? Describe the process you used to arrive at your conclusion.
• Write down the code you would use to include a 3rd variable, of your choosing, in your lm().

On your own

• Write down which other variables in the movie data you think would help you make better predictions.
  • Are there any variables that you think would not improve our predictions?
• Create a model for all of the variables you think are relevant.
  • Assess whether your model makes more accurate predictions for the testing data than the model that included only runtime and reviews_num.
• With your neighbors, determine which combination of variables leads to the best predictions for the testing data.