| Summary: | 碩士 === 國立臺灣大學 === 生醫電子與資訊學研究所 === 101 === Due to the progress of medicine, clinical data are increased very rapidly and biochemistry laboratory items are multiply measured with the subsequent consultations of patients. These multiple measurements clinical data may become another problem during analysis. This study proposes a practicable method to appropriately handle the clinical data with multiple measurements. Based on the case-based reasoning (CBR) method, we propose a multiple measurements CBR (MMCBR) method, extended from single measurement CBR (SingleCBR), for analyzing clinical data. The research target of this study is the prediction of recurrent status of liver cancer patients after receiving the first treatment in one year. We randomly separated dataset into four subsets, and the average results of classification using three-fold cross validation in four random datasets are analyzed, respectively. The results show models with better performance in the mean accuracy of four random datasets. Combination CBR could produce comparable results with SingleCBR and might have better stability than that of SingleCBR according to the standard deviation of accuracy. The mean sensitivities of MMCBR and Combination CBR in most combinations are better than those of SingleCBR. In this study, five feature selection approaches, different time periods of clinical data merging, and different weights are examined for establishing a predictive model.
|
|---|
