Feature selection algorithm ensembling based on meta-learning
Igor Tanfilev, Andrey A. Filchenkov, Ivan Smetannikov · 2017
We propose a new approach to feature selection that is based on ensembling and meta-learning. Meta-learning is used to choose feature selection algorithms from a preselected that are used to produce feature rankings, which are then aggregated into a resulting feature ranking. This approach requires a lot of additional computational time for meta-learning system construction, but it works fast and shows better feature selection quality than algorithms in aggregation.