机器学习中的多侧面递进算法MIDA

对高维海量数据,为解决准确率与泛化能力之间的矛盾,提出机器学习中的多侧面递进算法MIDA(Multi-side Increase by Degrees Algorithm),该算法将样本集分成几个部分,对各部分分别选择一组适应它们的特征子集.这种分而治之的方法,在保证一定的精度的前提下,符合人类对复杂问题的求解分重点,多方面考虑的方式,可有效地识别复杂问题的分类,提高泛化能力,降低了计算的复杂性.本文利用覆盖算法给出具体的多侧面递进算法,并给出实验结果,实验结果表明新的方法是有效的.

A Multi-Side Increase by Degrees Algorithm at Machine Learning

The conflict between validity and extensibility can be solved by using a multi side increase by degrees algorithm(shortened form MIDA) at machine learning in a data set with a feature space of high dimensionality and with large amount of samples that belong to many different classes.In the algorithm,the sample set is divided into several sample subsets step by step by double point.These feature subsets to match each sample set may be extracted at the same time.The method of different treatment to each sample subset is similas to that facing difficult problems people consider and seek a answer from different emphases and multi sides.The algorithm can classify the difficult problems effectually and raise the extensibility and reduce the complexity in condition of established accuracy.The multi side increase by degrees algorithm bases on a covering algorithm at machine learning.MIDA is used to classify a date set from Shanghai's stock,and the result is satisfied.