基于数据挖掘技术的黄土分类问题研究

依据数据挖掘技术,采用分类回归树决策树和概率神经网络对黄土的分类规则进行挖掘。利用主成分分析法对数据进行了清洗和降维处理,以处理后的新变量作为挖掘对象,使挖掘出的分类模型和规则得到了简化,提高了计算精度;同时归纳出了影响黄土分类的因素,所挖掘出的分类规则可用于黄土地层的智能划分。研究结果表明,挖掘出的知识具有良好的实用性。

LOESS CLASSIFICATION USING LOESS MECHANICAL DATA MINING SYSTEM

Data mining(DM) is a new information technology and a key knowledge discovery in database mining.It can process knowledge and information from a lot of practical data with incompletion,noise,fuzzy and uncertainty.Based on DM theory,the loess mechanical data mining system(LMDMS) was developed;and LMDMS was applied to the classification loess mechanical properties.The classification and regression trees(CART) decision trees and probabilistic neural network(PNN) in the LMDMS were applied to mining the loess classification rules and the principal component analysis was applied to compressing data to reach reducing dimension.After data were processed by the principal component analysis,the new variables could be obtained from mined objectives.Through the analysis of engineering application,the results indicate that:(1) the loess mechanical properties can be thoroughly seen by using the LMDMS based on a lot of loess mechanical basis physical indexes in practical engineering;(2) through comparing an algorithm of CART decision trees with an algorithm of probabilistic neural network in loess classification,an algorithm of CART decision trees is simpler in computation methodf,aster in computation velocity,and higher in computation precision(;3) CART decision trees method and probabilistic neural network are applied to mining loess classification rules and to constructing PNN model for loess classification,respectively,Meanwhile,coupling CART decision trees method and probabilistic neural network model can intelligently classify loess strata for loess engineering application;and(4) the principal component analysis can compress data,reduce data dimension,and simplify model.Meanwhile,it can improve computation velocity and precision and PNN model.The achieved results show that proposed model and rules are effective in engineering practices.