5 Machine Learning for IoT
293
Fig. 5.47 An overview of bootstrap aggregating based on random sampling with replacement
technique
estimator. Finally, the outputs of these weak models are combined by voting or
simple averaging to provide a meta-model (see Fig. 5.47).
5.4.7.2 Random Forest
Random forest (RF) or random decision forest is one of the implementations of
the bagging technique. Simplicity, flexibility, and great results have made RF one
of the most widely used algorithms for machine learning. The term forest points
to an ensemble of multiple decision trees (each of them is called estimator). This
ensemble of decision trees (estimator) is merged in the random forest method to
obtain predictions with higher accuracy.
That being said, RF adds additional randomness to bagging model by modifying
the original decision tree. Recall that in the original decision tree, in each iteration,
we need to select an attribute/feature to split upon. This attribute is selected among
all available attributes. However, in the modified learning algorithm, when we want