Support vector machine for solving imbalanced dataset problem
Most of machine learning classifiers such as Neural Network (NN), Naïve Bayes, and Decision Tree Method C4.5 are failed to classify the data when it deals with imbalanced data set. This is because; most of classifiers are biased to the majority class, tend to ignore minority class and treat the mino...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2012
|
| Subjects: | |
| Online Access: | http://eprints.utm.my/id/eprint/32546/1/IsmailMohdKhairuddinMFKE2012.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Most of machine learning classifiers such as Neural Network (NN), Naïve Bayes, and Decision Tree Method C4.5 are failed to classify the data when it deals with imbalanced data set. This is because; most of classifiers are biased to the majority class, tend to ignore minority class and treat the minority class as a noise/disturbance/variance. Generally, to tackle the imbalanced data set problem it consists of two strategies which are data level and algorithm level. The data level method consist of features selection and re-sampling the data such as undersampling, oversampling and combination of both undersampling and oversampling, while for algorithm level it consist internal modification of learning program. In this project, the Support Vector Machine (SVM) classifier is proposed in order to investigate the imbalanced data set problem. The investigation is obtained by measured the performance based on SVM classifier. This investigation will cover and measure the performance SVM classifier by measuring the g-mean value. The performance of SVM classifier is measured by measuring the g-mean value .Therefore, in order to increase the performance of SVM classifier oversampling methods called SMOTE is introduced and combine with it and the g-mean value is calculated. Experimental validation on the proposed algorithm is performed and demonstrated on various set of imbalanced data sets. Some experiment have been design to validate the proposed algorithm and performed it with various set of imbalanced data sets. Finally, the result is for each proposed algorithm is being compared and analyze. |
|---|