machine learning - Values greater than 1 in svm prediction file -

i using svm light train model binary classification. using model, tested examples. surprised see output of prediction file, contains values greater 1 less -1. thought range [-1,1]. doing wrong?

it makes sense why values not bounded interval of [-1, 1] if understand how svm works. svm tries draw line separates negative , positive data points while maximizing distances line.

the values in prediction file represent distances of data svm optimal hyperplane, positive values on positive class side of hyperplane , negative values on negative class side of hyperplane. these distance can arbitrarily large or small , not bounded can seen image:

i've seen svm implementations such weka's implementation of platt's smo normalize values confidence values on positive class bounded interval of [0, 1], both ways work fine determining how confident svm on classification since data point further hyperplane more 1 lying close hyperplane.