Detecting olfactory impairment through objective diagnosis: Catboost classifier on EEG data
- Authors
- Cheon, Min-jong; Lee, Ook
- Issue Date
- Jul-2021
- Publisher
- Little Lion Scientific
- Keywords
- Artificial Intelligence; Deep Learning; Diagnosis; EEG; Machine Learning; Olfactory Impairment
- Citation
- Journal of Theoretical and Applied Information Technology, v.99, no.14, pp.3596 - 3604
- Indexed
- SCOPUS
- Journal Title
- Journal of Theoretical and Applied Information Technology
- Volume
- 99
- Number
- 14
- Start Page
- 3596
- End Page
- 3604
- URI
- https://scholarworks.bwise.kr/hanyang/handle/2021.sw.hanyang/141486
- ISSN
- 1992-8645
- Abstract
- Detecting olfactory impairment using an objective diagnosis kit has been a challenge. Recently, machine learning and deep learning models have been used on EEG data with promising results. The goal of our study was to detect olfactory impairment through a machine learning classifier with EEG data. This was done by identifying the important EEG data factors affecting olfactory impairment. Finally, we compared our model to other machine learning and deep learning algorithms in order to identify possibilities for further research. Downsampling and extracting various waves from EEG data were conducted for data preprocessing. Then, an independent component analysis was performed to remove artifacts. Through this processing, a dataset in CSV format was obtained. Next, we built a CatBoost classifier model because it is recent boost model and has high performance for classification. It identified whether a subject had olfactory impairment or not. After training with the CatBoost algorithm, we compared it to different machine learning and deep learning algorithms. The CatBoost model showed 87.56 % accuracy, while other machine learning algorithms such as the random forest classifier, gradient boosting classifier, XG boosting classifier, k-nearest-neighbor classifier, decision tree classifier, Gaussian NB, and logistic regressor revealed 82.22 %, 78.89 %, 78.22 %, 75.78 %, 74 %, 69.78 %, and 41.11 % accuracy, respectively. With deep learning models, which consisted of bi-directional long short term memory, long short term memory and a deep neural network, the performance was 63.11 %, 51.33 %, and 60 %. The CatBoost model showed feature importance, which revealed that the gamma wave on the Cz channel was about 20, which was the highest among the other variables.
- Files in This Item
-
Go to Link
- Appears in
Collections - 서울 공과대학 > 서울 정보시스템학과 > 1. Journal Articles
![qrcode](https://api.qrserver.com/v1/create-qr-code/?size=55x55&data=https://scholarworks.bwise.kr/hanyang/handle/2021.sw.hanyang/141486)
Items in ScholarWorks are protected by copyright, with all rights reserved, unless otherwise indicated.