Feature Analysis and Machine Learning for Migraine Detection

• Main Authors: Lin, Yu-Chian, 林宇謙
• Other Authors: Tien, Chung-Hao
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/qz2ft4

碩士 === 國立交通大學 === 光電工程研究所 === 107 === This study aimed to establish a migraine predictive model by analyzing the iris color. A total of 339 "Iris Images" were collected from 269 migraine patients and 70 controls. To analyze iris image we chose twelve color features, including original color space RGB, brightness, chrominance color space YCbCr, CIE L*a*b*, and HSV cylindrical coordinate color space. Additionally, to improve accuracy of detection, we add five iris texture features (PC1~PC5) to each iris image (10×10 pixel) using principal component analysis (PCA). Regarding migraine predictive model, our study showed the most suitable algorithm was logistic regression by using probability model method in machine learning. The accuracy rate of logistic regression was compared with that of decision tree, support vector machine, k nearest neighbors, random forest, Naive Bayes classification, and neural network. Overall, the accuracy of our migraine predicting model was 76.5% using logistic regression method on twelve color features and five iris texture features of iris image.