Multiple Learners Combination: Stacking

Advanced computational techniques have found ever increasing applications in bioinformatics. In this field, machine learning techniques applied to prediction problems are of particular relevance. Of course, the more accurate the models are, the better the prediction will be. However, for free domain problems, theoretical evidence shows that there is no predictor always having better performance than another. The ensemble methods arise from the idea that instead of looking for the learning algorithm inducing the “optimal” learner for that particular problem, accuracy can be improved by combining more “weak” learner to obtain a “strong” one. This article presents stacking, a general method for combining multiple learner that can be approached to cross-fold validation.