玻璃文物特征分析及风化预测模型——基于机器学习与统计分析

Characteristic Analysis and Weathering Prediction Model of Glass Cultural Relics—Based on Machine Learning and Statistical Analysis

摘要: 本文基于2022年高教杯C题，首先将数据集行one-hot编码，对化学成分数据进行填补缺失值，并使用Person相关系数对数据相关性进行分析。后基于奇异值分解的PCA算法对数据降维。通过随机森林、支持向量机、Xgboost、Logistic回归对降维后的数据进行分类并求得决策边界，基于软分类器预测出了各个文物的风化程度，得到在文物信息数据的验证集分类准确度达到了86.7%，化学成分数据达到了94.1%。之后利用召回率、f1值、精度、ROC曲线等评价指标对模型进行了评价与选择。最终得到了各特征的相关性以及各文物的风化概率，同时得出了Xgboost在该数据集中预测的优越性。

Abstract: Based on the 2022 Higher Education Cup A question, this paper first codes the data set line one-hot, fills in the missing values for the chemical composition data, and analyzes the data correlation using the Person correlation coefficient. After the PCA algorithm based on singular value decomposition reduces the dimensionality of the data. Through random forest, support vector machine, Xgboost, logistic regression to classify the dimensionality reduction data and find the decision boundary, based on the soft classifier, the degree of weathering of each cultural relics is predicted, and the classification accuracy of the verification set of the cultural relics information data reaches 86.7%, and the chemical composition data reaches 94.1%. After that, the model was evaluated and selected using evaluation indicators such as recall rate, f1 value, accuracy, and ROC curve. Finally, the correlation of each feature and the weathering probability of each artifact are obtained, and the superiority of Xgboost’s prediction in this data set is obtained.