R Tutorial: Multivariate GAMs

Want to learn more? Take the full course at https://learn.datacamp.com/courses/nonlinear-modeling-in-r-with-gams at your own pace. More than a video, you'll learn hands-on coding & quickly apply skills to your daily work. --- So far, all the GAMs we have seen have been univariate; they have a single predictor for the outcome. However, we can perform multiple regression with GAMs. Multiple GAMS also can contain a mixture of smooths, linear effects, and continuous or categorical variables. In this lesson, we'll learn how to use this flexibility to fit a variety of different models to data. We'll now work with the mpg data set. This is a data set of 205 models of cars, consisting of various traits like their make, model, cylinders, price and weight, and their city and highway fuel efficiency. We'll be building models that use the vehicle traits to predict fuel efficiency. Let's start with a very simple one variable model. Here is the code for a model that predicts highway fuel efficiency as a smooth function of automobile weight. The resulting model captures the nonlinear decreasing relationship between these two variables. To add an additional variable, such as length, we just include another s() function in our formula, separated by a plus sign. Here, we add car length as another predictor. We see from these plots that length has increasing nonlinear effect on fuel economy, and this effect is weaker than the weight effect. Note that, in this model, both the effect of weight and price are non-linear terms, but the two are simply added together to get a final prediction. That addition is where the additive in generalized additive models comes from. Not every term in a GAM has to be nonlinear. You can combine linear and nonlinear terms. To add a linear term, don't wrap the predictor in the s() function. Here I've made the length term from the previous model linear. In practice, we rarely make continuous variables linear in GAMs. This is because, if the