International Conference on Computer and Knowledge Engineering

Home / 11th International Conference on Computer and Knowledge Engineering

Improving the classification of high dimensional class-imbalanced data using the Chaos particle swarm optimization with Levy Flight

Authors :

Mohammad Ali Zarif¹ Javad Hamidzadeh²

1- Sadjad University of Technology 2- Sadjad University of Technology

Keywords :

Imbalanced data classification; High Dimensional data; Cost-sensitive Learning; Feature Selection; Chaos Particle Swarm Optimization; Levy Flight

Abstract :

Nowadays classification of imbalanced datasets has gone under considerable attention in the field of data mining and machine learning. Classification of imbalanced data is a challenging task due to imbalanced distribution of the data. In addition, conventional algorithms do not perform well in classification of such high dimensional data. Researchers have proposed many techniques for improving the classification of high dimensional class-imbalanced data. These approaches fall into five categories: 1- Data level, 2- Algorithm level, 3- Cost-sensitive learning, 4- Ensemble learning and 5- Feature selection methods. In this paper, we have exploited cost-sensitive and feature selection methods. Our proposed method first uses fuzzy c-means clustering (FCM) to estimate the probability of each sample being a member of the minority class. Using this method, samples which are hard to classify are determined based on their probabilities and then the cost-sensitive learning is employed to estimate their cost. Finally, particle swarm optimization algorithm enhanced by levy flight is applied for feature selection and a SVM model is trained for classification of the data. By using particle swarm optimization alongside with SVM classifier, best features are determined and misclassification rate of minority samples is reduced. The evaluation of the proposed method is done by using G-Mean metric and 10-fold cross validation. The experimental results demonstrate the superiority of the proposed method compared to other approaches in classification of high dimensional class-imbalanced data.