Machine Learning for the Preliminary Diagnosis of Dementia

Algorithm 1

The Random Forest algorithm for feature selection.

	Input: A training set: , , ,
	where n is the size of the training set, denotes the features in the sample, denotes the class label in the sample, and X denotes the feature space
	Output: The key feature T;
	Begin
(1)	Set all the feature weights is 0, T is empty;
(2)	for i = 1 to m do;
(3)	Given a tree ensemble model
(4)	Computes the importance of each feature.
	Average over several randomized trees:
	Importance (feature t) = sum (over nodes which split on feature t) of the gain, where gain is scaled by the number of instances passing through node,
	Normalize importance for tree to sum to 1.
	Normalize feature important vector to sum to 1.
(5)	T = the intersection of the set of the set of .
	End