# 9 Model averaging approaches

All the techniques presented so far require a model selection procedure where different model structures are assessed and compared in order to attain the best representation of the data. In model selection the winner-takes-all approach is intuitively the approach which should work the best. However, recent results in machine learning show that the performance of the final model can be improved not by choosing the model structure which is expected to predict the best but by creating a model whose output is the combination of the output of models having different structures. The reason is that, in fact, every hypothesis $h(\cdot ,\alpha_N)$ is only an estimate of the real target and, like any estimate, is affected by a bias and a variance term. The theoretical results of Section 3.10 show that a variance reduction can be obtained by simply combining uncorrelated estimators. This simple idea underlies some of the most effective techniques recently proposed in machine learning. This chapter will sketch some of them.