Title: An oversampling technique using genetic algorithm and feature selection for imbalance learning

Authors: Peter Awonnatemi Agbedemnab; Moses Apambila Agebure; Abaafe Awonate Emmanuel

Addresses: Department of Information Systems and Technology, C.K. Tedam University of Technology and Applied Sciences, Navrongo, Ghana ' Department of Computer Science, C.K. Tedam University of Technology and Applied Sciences, Navrongo, Ghana ' Department of Computer Science, C.K. Tedam University of Technology and Applied Sciences, Navrongo, Ghana

Abstract: This study presents an oversampling technique that improves the performance of traditional classification algorithms when learning from imbalance datasets. The proposed technique focuses on passing on the best features within the dataset to the next generation by leveraging on feature selection, and the selection and crossover operators of genetic algorithm. The study used Euclidean distance as the fitness function and the feature selection as a metric for the crossover function. We used machine learning classification algorithms from the scikit-learn library namely, random forest and decision tree on six datasets with different imbalance ratios for test runs. The performance of the proposed technique is evaluated against SMOTE, ROS, RUS and when no sampling is done (NONE). Using area under the curve and geometric mean metrics, it was observed that there is a significant improvement of the performance of the models when the proposed technique is used for oversampling before training.

Keywords: crossover operator; Euclidean distance; genetic algorithm; imbalance learning; oversampling.

DOI: 10.1504/IJAISC.2024.146628

International Journal of Artificial Intelligence and Soft Computing, 2024 Vol.8 No.4, pp.339 - 355

Received: 21 Mar 2024
Accepted: 19 Nov 2024
Published online: 09 Jun 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article