Relief algorithm is an algorithm of feature selection.
suppose we have a data set with the count of and features, each feature is scaled to the interval [0,1] (0 and 1 for binary data), and the data set belongs to two known classes.
for the th feature, we calculate the weight for the feature by the following:
where is the closest instance belonging to the same class of instance , and is the closest instance belonging to the different class of instance . the feature weight bigger than a threshold is relevant feature.
网友评论