Siavash, Are you comparing your test results with someone elses? If so, t-test is one of many tests for determining whether or not the differences seen in one's score are meaningful. Much of machine learning omits statistics, so it not mandatory.
Daniel. On Mon, Apr 4, 2011 at 5:58 AM, SIAVASH GHODSI MOGHADDAM <[email protected]> wrote: > > Dear Mahout Friends, > In my research (My master thesis), I am using some Clustering techniques to > cluster Users and then I apply CF algorithm on these clusters and finally I > evaluate the accuracy by using MAE. > I use MovieLens 100K dataset, which is already disjoint for 5 fold cross > validation, I cluster each fold training data, apply the CF and average all > the clusters MAE in this fold and then do the cross validation on all folds. > My question is: Do I need t-test? > I seldom see in the literature that anyone use t-test for recommendation > systems significant test, except “Clifford Lyon” on his final > project<http://materialobjects.com/cf/MovieRecommender.pdf>! > Therefore, I am quite bluer whether I should do the t-test or not! > I would highly be appreciated if you tell me your idea. > > Thanks in advance, > Siavash > University Technology Malaysia<http://www.utm.my/> >
