基于集成特征选择和随机森林的古代玻璃分类模型

Classifying Model of Ancient Glass Products Based on Ensemble Feature Selection and Random Forest

将机器学习中的特征选择方法和分类算法融入古代玻璃制品成分分析和类别鉴定问题研究,以准确率和AUC作为分类性能度量指标,尝试构造古代玻璃制品化学成分选择的集成特征选择模型和鉴定分类的随机森林模型.对不同特征选择方法的结果进行集成,选择重要的化学成分,对选出的重要特征结合随机森林模型、k近邻学习和Naive Bayes模型等方法进行实验分析.结果表明,采用集成特征选择分析出氧化铅、氧化钡、氧化钾等成分对玻璃表面风化影响比较显著,且高钾玻璃中这3种成分两两关联很大,对选出的重要特征应用基于k折交叉验证的随机森林模型进行分类得到的准确率较高,模型稳定性强.该方法可以为我国古代玻璃制品的成分分析和类别鉴定提供理论参考,对其它玻璃的相似研究也有一定程度的借鉴意义.

This paper was to construct ensemble feature selection and random forest for the identification of ancient glass products via integrating the machine learning algorithm into the identification and analysis of ancient glass products and taking accuracy rate and AUC as the measurement indexes of classification performance.The results of different feature selection methods were analyzed,and the important chemical components were selected.The selected important features with random forest,k-nearest neighbor learning and naive Bayesian were investigated.The results show that lead oxide,barium oxide,and potassium oxide have an impact on the weathering of the glass surface by using ensemble feature selection.In high potassium glass,three components are closely related,the accuracy of classification by the random forest based on the k-fold cross validation for selected important features is great,and the model is stable.This method can provide a theoretical reference for the composition analysis and category identification of ancient glass products,and otherglasses.