A Study on AdaBoost Algorithm

• Main Authors: Hsieh, Chih-Yu, 謝志優
• Other Authors: Chen, Chaur-Chin
• Format: Others
• Language: en_US
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/18874159480075931972

碩士 === 國立清華大學 === 資訊系統與應用研究所 === 102 === Classification is one of the technology in data mining, we can discover patterns and relationships between parameters in data by classification that we can use these patterns to predict unknown data. In the real life, it is applied in several areas. For example, we can discover patterns from the genes of patients by using classification and then it can apply to other patients by using this pattern. Thus, data mining is the most important technology in data analysis. In Big Data, it cannot obtain the information without using data mining. In this thesis, we study the binary and multiclass classification of AdaBoost algorithm. In this algorithm, each sample has a weight value. It uses T weak classifiers to train the training samples. In training weak classifiers, we must change the weight of each incorrectly and correctly classified sample. Finally, the strong classifier is to combine the votes of all individual weak classifiers and then we can use this strong classifier to predict the unknown data. Experiments on colon cancer, breast cancer, 8OX, and Iris data sets are illustrated.