微阵列数据中一种改进Bagging决策树算法的研究

On an Improved Bagging Decision Tree Algorithm in Microarray Data

针对基因微阵列数据具有高维度、小样本等独特的特点,本文研究并实现了旨在降低计算时间和提高精确度的Bagging决策树。本文提出了一个能极大地降低计算时间、同时对精确度影响不大的属性离散化过程,接着以一种新的类分布置信度的方式构造决策树,该方法在最终的Bagging组合方面有一定的优势。结合上述方法的Bagging决策树算法在基因微阵列数据集分类上取得了良好的效果。

Based on the characteristics of high dimension and small sample, this paper investigates the improvements of bagging decision trees which aim mainly at improving computation time and accuracy. A discretization procedure is proposed, resulting in a dramatic speedup without significant decrease in accuracy. Then a new class distribution confidence is suggested improving the accuracy of the final bagging decision tree. Combining these improvements makes it get excellent performance on gene microarray data.