摘要
在生物统计以及数据挖掘中,分类预测是最基本的任务之一。本文将探讨一种新的方法—助推偏最小二乘法(BPLS)。它结合了一系列收缩的偏最小二乘模型,每个模型只有一个主成分。这种新方法和传统的偏最小二乘方法不同,它不需要选择一系列的偏最小二乘成分。只需要确定两个参数即可。通过对真实数据的训练,得以证明这种新方法比传统的偏最小二乘法在防止过度拟合方面有更好的表现,同时能够保证精确度。
Classification is one of the most basic tasks during the biological statistics and data mining. Tiffs article will explore a new method - PLS - propelled (BPIS).It combines a series of models, which is partial least - square,and each roodel is only one principal component. This new method is different from the traditional one since it need not to choose a series of partial least square component, and it only need to determine two parameters. In this paper, by training seven groups of real data, we prove that this new one has better performance in guarding against over- fitting, and ensures the accuracy at the same time.
出处
《数学理论与应用》
2009年第4期118-121,共4页
Mathematical Theory and Applications
关键词
数据挖掘
分类预测
偏最小二乘
Data mining Classification Partial least squares