期刊论文

Algorithm improvement based on random forest trip recognition technology

中文

武汉轻工大学学报

2095-7386

2018

37

3

74-80

10.3969/j.issn.2095-7386.2018.03.013

本校为第一机构

随机森林是一种组合分类算法,随机森林组合分类器具有较强的数据集适应能力,无需进行繁琐的数据规范化处理,比较适合处理多类别的分类问题。在基于智能手机的出行方式识别系统中,智能手机对源数据的预处理十分有限,而随机森林算法抗噪声能力较强,因此随机森林算法的抗噪声性能十分关键。但对于随机森林算法而言,由于决策树算法容易发生过度拟合问题,因此对于出行方式识别也存在一定的劣势。本文针对随机森林分类器进行算法对比,并从提高少数类选中概率的角度进行算法优化,能够在一定程度上保证训练集特征的全面性,对于改进随机森林算法性能具有一定的意义。

Random forest is a combination of classification algorithm, and random forest combination classifier has a strong ability to adapt to the data set, without cumbersome data normalization, more suitable for handling multi-class classification problem. In the smart phone-based travel mode recognition system, the training efficiency and prediction ability of the algorithm are demanding. Stochastic forest combiner has excellent anti-noise ability. Gen-erally speaking, for the data collected by smart phones, the source data pre-processing is very limited, so the anti-noise performance of random fore...