International Conference on Computer and Knowledge Engineering

Home / 11th International Conference on Computer and Knowledge Engineering

Instance Selection from Skewed Class Distributions by Using the multi-objective optimizer

Authors :

Mona Moradi¹ Javad Hamidzadeh²

1- university of Semnan 2- Sadjad University

Keywords :

Instance selection, Combined fuzzy weighted average distance-based decision surface, Chaos-based firefly algorithm, Convergence speed, Classification accuracy, Reduction rate.

Abstract :

Instance selection deals with reducing the dataset size by removing as much uninformative and unrepresentative data as possible. The reduced dataset would be used as a training set of any classifier, so it has a direct influence on the performance of a classifier. Two issues to consider during instance selection are: (1) the amount of dataset reduction and (2) conservation of generalization performance. Since maximizing dataset reduction leads to poor performance, both issues conflict. Therefore, instance selection is intrinsically a multi-objective problem. Instance selection becomes complicated in skewed distributions. The proposed method resolves this issue by a chaos-based evolutionary algorithm and keeps only the boundary instances by a combined fuzzy weighted average distance-based decision surface. The performance has been evaluated on real-world datasets by the 10-fold cross-validation method. Evaluations manifest the competitive performance for instance selection in terms of error rate, reduction rate, and Gmean.