Applying Hybrid Data Preprocessing Methods in Stroke Prediction

• Main Authors: Muhammad Rieza, 慕哈曼
• Other Authors: Chao Ou-Yang
• Format: Others
• Language: en_US
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/32906847799465722877

碩士 === 國立臺灣科技大學 === 工業管理系 === 101 === Stroke has always been highlighted as a big threat of health in the worldwide. Brain image examination and ultrasound are some alternatives to discover stroke disease. Data mining has been used widely in many areas, including medical industry. The uses of data mining methods allow doctors to make prediction of certain diseases. Therefore, in this research, hybrid model integrating imbalance data preprocessing, feature selection, and back propagation network, support vector machine, decision tree for stroke prediction. The dataset used is brain examination data which collected from 2004 to 2011. However, highly imbalance dataset available can impact the performance of prediction as well as feature selected. The study firstly “rebalance” the dataset by comparing sampling methods; Random Under Sampling by Age and RUSboost. In addition, important features of balance training dataset would be selected by information gain and stepwise regression analysis. Towards the end, selected features would be processed using Back Propagation Network, Support Vector Machine and Decision Tree to predict the stroke. These hybrid methods may assist doctor to provide some possibilities information to the patient.