Linear regression with principal components
The object newsData now contains an additional variable: logShares. The number of shares tell you how often the news articles have been shared. This distribution, however, would be highly skewed, so you are going to work with the logarithm of the number of shares. Apply what you just learned and predict the log shares!
This exercise is part of the course
Machine Learning for Marketing Analytics in R
Exercise instructions
- Compute a model to predict the log shares with all other variables. Store it as
mod1. - Create a new dataframe
dataNewsComponentswith the log shares and the values on the first 6 components. The objectpcaNewsagain contains the PCA results. - Compute a second model (
mod2) that predicts the log shares with just the 6 components. - Compare the adjusted R squared of the models. How did the value change by using only the principal components? How good is your model?
Hands-on interactive exercise
Have a go at this exercise by completing this sample code.
# Predict log shares with all original variables
mod1 <- lm(logShares ~ ., data = ___)
# Create dataframe with log shares and first 6 components
dataNewsComponents <- cbind(logShares = newsData[, "logShares"],
___$x[, 1:__]) %>%
as.data.frame()
# Predict log shares with first six components
mod2 <- lm(___ ~ ., data = ___)
# Print adjusted R squared for both models
___(mod1)$adj.r.squared
summary(___)$___