Filter-based optimization techniques for selection of feature subsets in ensemble systems

Classification of Implantable Rotary Blood Pump States With Class Noise

“…The advantages of wrapper-based methods include that they take into account the ability of a classifier model to approximate or accommodate the distribution of a proposed feature [37], hence the final selected feature subset, when used with the desired classifier, often performs better than features selected by a filter-based method [35]. The main limitation of wrapper-based methods is that they are computationally intensive, which can be an issue when the data set contains a large number of features [38]. Compared with filter-based methods, they are susceptible to a higher risk of over-fitting, or require that additional data is acquired with which to perform final validation of the trained classifier model [36].…”

“…Compared with filter-based methods, they are susceptible to a higher risk of over-fitting, or require that additional data is acquired with which to perform final validation of the trained classifier model [36]. In addition, wrapper-based feature selection must be re-run each time a new classifier model is used [38].…”

Genetic Algorithm for the Column Subset Selection Problem

“…Recently, Santana and Canuto [16] solved a slightly different problem of the selection of a subset of features for ensemble systems. The authors used bio-inspired algorithms including Ant Colony Optimization, Particle Swarm Optimization, and Genetic Algorithms to solve the problem and the experiments showed that all three algorithms were able to solve that problem very efficiently.…”

Prediction approach of software fault-proneness based on hybrid artificial neural network and quantum particle swarm optimization

“…At present, for the identification of relevant metrics and the removal of irrelevant metrics, three different methods can be employed, namely the filter [22,23], wrapper [24,25], and hybrid models [26,27]. The filter model relies on general characteristics of the data to evaluate and select metrics subsets without involving prediction algorithm.…”