Two-stage individual gene selection methods based on misclassification cost and accuracy

• Main Authors: Jui-YangHsu, 許瑞洋
• Other Authors: Tzu-Tsung Wong
• Format: Others
• Language: zh-TW
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/16808233634669487228

碩士 === 國立成功大學 === 資訊管理研究所 === 98 === As the continuous improvement and innovation of bioscience and medical science technologies, scientists developed a series of technologies for gene microarray data to find the relations between diseases and genes. The number of instances in a microar-ray data set is far less than the number of genes in an instances, and lots of genes are irrelevant to a specific disease. Therefore, gene selection is essential to reduce the di-mensionality of a microarray data set. The misclassification costs of different classes are generally different. Previous study performs cost-sensitive gene selection such that the classification accuracy of a microarray data set is greatly reduced. To com-pensate such difficiency, this study considers both misclassification cost and predic-tion accuracy to propose 18 two-stage individual gene selection methods for microar-ray data. The experimental results on eight microarray data sets show that the method adopting probability ranking for misclassification cost in the first stage and t-value ranking for prediction accuracy in the second stage has the best performance evalu-ated by area under cost curve and area under accuracy curve.