Feature selection for multi-purpose predictive models: a many-objective task

Authors:
Alan P. Reynolds;David W. Corne;Michael J. Chantler
Affiliations:
School of Mathematical and Computer Sciences, Heriot-Watt University, Edinburgh, Scotland;School of Mathematical and Computer Sciences, Heriot-Watt University, Edinburgh, Scotland;School of Mathematical and Computer Sciences, Heriot-Watt University, Edinburgh, Scotland
Venue:
PPSN'10 Proceedings of the 11th international conference on Parallel problem solving from nature: Part I
Year:
2010

Citing 5
Cited 1

A Fast Elitist Non-dominated Sorting Genetic Algorithm for Multi-objective Optimisation: NSGA-II

PPSN VI Proceedings of the 6th International Conference on Parallel Problem Solving from Nature
Feature Selection Using Multi-Objective Genetic Algorithms for Handwritten Digit Recognition

ICPR '02 Proceedings of the 16 th International Conference on Pattern Recognition (ICPR'02) Volume 1 - Volume 1
An introduction to ROC analysis

Pattern Recognition Letters - Special issue: ROC analysis in pattern recognition
Data Mining: Practical Machine Learning Tools and Techniques, Second Edition (Morgan Kaufmann Series in Data Management Systems)

Data Mining: Practical Machine Learning Tools and Techniques, Second Edition (Morgan Kaufmann Series in Data Management Systems)
The use of the area under the ROC curve in the evaluation of machine learning algorithms

Pattern Recognition

Multi-objective feature selection in music genre and style recognition tasks

Proceedings of the 13th annual conference on Genetic and evolutionary computation

Quantified Score

Hi-index	0.00

Visualization

Abstract

The target of machine learning is a predictive model that performs well on unseen data. Often, such a model has multiple intended uses, related to different points in the tradeoff between (e.g.) sensitivity and specificity. Moreover, when feature selection is required, different feature subsets will suit different target performance characteristics. Given a feature selection task with such multiple distinct requirements, one is in fact faced with a very-many-objective optimization task, whose target is a Pareto surface of feature subsets, each specialized for (e.g.) a different sensitivity/specificity tradeoff profile. We argue that this view has many advantages. We motivate, develop and test such an approach. We show that it can be achieved successfully using a dominance-based multiobjective algorithm, despite an arbitrarily large number of objectives.