EKC2023

[P41-LH/EI]A Study on Various Oversampling Methods for Voice Pathology Detection Using the Pathological Database of Eulji Medical Center

Not scheduled

20m

Poster Poster(Wed)

Speaker

Prof. Ji-Yeoun Lee (Eulji University)

Description

This study investigates various oversampling methods for enhancing voice pathology detection using the pathological database of Eulji Medical Center. Voice pathology detection is crucial in diagnosing and treating speech disorders, but imbalanced datasets pose a challenge. The study utilizes the comprehensive pathological database from Eulji Medical Center, providing diverse voice pathology samples for analysis.
We evaluate and compare oversampling methods, including random oversampling, SMOTE, ADASYN, and borderline-SMOTE. These methods generate synthetic samples for minority classes, balancing the dataset. We analyze the impact of oversampling on voice pathology detection performance, considering accuracy, precision, recall, and F1-score.
Results show that oversampling effectively addresses class imbalance and improves voice pathology detection performance. We identify the most effective oversampling method for this task through comparative analysis. This study contributes to advancing voice pathology detection methods and provides insights for accurate diagnostic systems in speech pathology.

References

1. Fan, Z. et al. Class-Imbalanced Voice Pathology Detection and Classification Using Fuzzy Cluster Oversampling Method, Appl. Sci. 2021, 10, 3450-3470. [CrossRef]
2. He, H.; Bai, Y.; Garcia, E.A.; Li, S. ADASYN: Adaptive synthetic sampling approach for imbalanced learning, in Proc. of the 2008 IEEE International Joint Conference on
   Neural Networks, Hong Kong, China, 2008, 1322–1328. [CrossRef]
3. Dong, Y.; Wang, X.A New Over-Sampling Approach: Random-SMOTE for Learning from Imbalanced Data Sets, In Proc. of the International Conference on Knowledge Science, Engineering and Management, Dalian, China, 2011, 10–12. [CrossRef]
4. Douzas, G.; Bacaoa, F.; Last, F. Improving imbalanced learning through a heuristic oversampling method based on k-means and SMOTE, Information sciences 2018, 465, 1-20. [CrossRef]
5. Wei-Chao, L.; Chih-Fong, T.; Ya-Han, H.; jing-Shang, J. Clustering-based undersampling in-class imbalanced data, Information sciences 2017, 409-410, 17-26. [CrossRef]
6. Lee J-N, Lee J-Y. An Efficient SMOTE-Based Deep Learning Model for Voice Pathology Detection. Applied Sciences. 2023; 13(6):3571. https://doi.org/10.3390/app13063571