Publication: Classification of Benign Paroxysmal Positioning Vertigo Types from Dizziness Handicap Inventory using Machine Learning Techniques
0
0
Issued Date
2018
Resource Type
File Type
application/pdf
Other identifier(s)
2-s2.0-85060005138
Rights Holder(s)
มหาวิทยาลัยศรีนครินทรวิโรฒ
Bibliographic Citation
2018 International Conference on Intelligent Informatics and Biomedical Sciences, ICIIBMS 2018. (2018), p.209-214
Suggested Citation
Masankaran L., Viyanon W., Mahasittiwat V. Classification of Benign Paroxysmal Positioning Vertigo Types from Dizziness Handicap Inventory using Machine Learning Techniques. 2018 International Conference on Intelligent Informatics and Biomedical Sciences, ICIIBMS 2018. (2018), p.209-214. doi:10.1109/ICIIBMS.2018.8550002 Retrieved from: https://hdl.handle.net/20.500.14740/5786
Author(s)
Abstract
Benign Paroxysmal Positioning Vertigo (BPPV) is one of the causes of vertigo which extremely affects the daily life of patients. Different types of BPPV are treated in a different way. Physicians differentiate the BPPV types using nystagmus characteristics. However, some patients have unclear nystagmus, so their treatments are delayed due to the difficulty of diagnosis. Dizziness Handicap Inventory (DHI) is a tool to assess the severity of dizziness before a patient is diagnosed by a physician. The use of DHI can distinguish BPPV types which can help physicians decide what treatments would be suitable for patients. This research aims to study the ability of using DHI for diferrential diagnosis of Posterior canal-Benign Paroxysmal Positioning Vertigo (PC-BPPV) and Horizontal canal-Benign Paroxysmal Positioning Vertigo (HC-BPPV) via machine learning techniques. We used feature selection techniques and feature engineering to increase the power of machine learning algorithms. Random Forest, Support vector machine, K-Nearest Neighbor and Naïve Bayes were used to develop predictive models from DHI features that have statistically significant. Accuracy, precision, recall, and F1-score were used to evaluate the performance of each model. It reveals that F7+E23, age and P8 are the top three important features and the model using Gaussian Naïve Bayes is the best model to discriminate HC-BPPV and PC-BPPV with 73.91% accuracy, 66.67% precision, 80.00% recall and 72.73% F1-score. In conclusion, the models that were created from DHI score can predict BPPV types at a certain level, but still not very good. Physicians have to use patient°s medical history and nystagmus observation for diagnosis. In the future, if we can collect more data or features, we may reduce the overfitting problem and have a better performance model. © 2018 IEEE.