Random forests are a type of bagging, with every tree in the forest potentially constructed in parallel by sampling the dataset with replacement.

The subsampling process helps induce variability in the model, thereby help reduce the overfitting problem seen with decision trees.