ML Activity

Your task: build a model to predict rates of racial segregation at the county level. First, choose your variables and split data into training and test sets. Using a random forest model, you must choose three variables as features (independent variables) in your model. Create a plot(s) to show the importance and node purity of the features in your model. Finally, check the rmse of the training and validation models to inspect for overfitting.

Warning message:
In normalizePath(quartoSrcFile, winslash = "/") :
  path[1]="": No such file or directory

model_data_rs <- data %>% 
  select(c("racial_segregation", "unemp_rate", "frac_middleclass", "poverty_rate"))

Error in data %>% select(c("racial_segregation", "unemp_rate", "frac_middleclass",  : 
  could not find function "%>%"

varImpPlot(rf.fit)

model.valid_rs$predict <- predict(rf.fit, newdata = model.valid_rs)

model.valid_rs <- model.valid_rs %>% 
  mutate(error = racial_segregation-predict) %>% 
  mutate(square_error = error^2) 

sqrt(mean(model.valid_rs$square_error, na.rm=T))

model.train_rs$predict <- predict(rf.fit, newdata = model.train_rs)

model.train_rs <- model.train_rs %>% 
  mutate(error = racial_segregation-predict) %>% 
  mutate(square_error = error^2) 

sqrt(mean(model.train_rs$square_error, na.rm=T))

#repeat this analysis using linear regression

lm1 <- lm(racial_segregation ~ unemp_rate+frac_middleclass+poverty_rate, data = model.train_rs)

summary(lm1)

model.valid_rs <- model.valid_rs %>% 
  mutate(yhat = predict(lm1, newdata = model.valid_rs)) %>% 
  mutate(lm_residual = racial_segregation - yhat) %>% 
  mutate(lm_residual_sq = lm_residual^2)

ggplot(model.valid_rs) +
  geom_point(aes(x=yhat, y=lm_residual)) +
  geom_hline(aes(yintercept=0), linetype="dashed", color="red") +
  xlab("Predicted Inequality") +
  ylab("Residual Inequality")+
  theme_minimal()+
  labs(title = "Residual Plot for Validation Model")

#calculate rmse

sqrt(mean(model.valid_rs$lm_residual_sq, na.rm=T))