基于集成分类的数据库指纹综合评价  

Comprehensive evaluation of database fingerprints based on ensemble classification

作　　者：李鼎文 张迎周[1] LI Dingwen;ZHANG Yingzhou(School of Computer Science,Nanjing University of Posts and Telecommunications,Nanjing 210023,China)

机构地区：[1]南京邮电大学计算机学院,江苏南京210023

出　　处：《南京邮电大学学报（自然科学版）》2023年第1期86-95,共10页Journal of Nanjing University of Posts and Telecommunications：Natural Science Edition

摘　　要：为了在给定的数据库及特定的使用场景下选取性能更高的数据库指纹,同时完善数据库指纹理论体系,对数据库指纹综合评价进行研究具有强烈的现实意义。针对数据库使用场景,结合传统数字指纹的性能评价指标,建立数据库指纹层次评价指标体系,并提出指标量化和评测数据预处理的解决方案。指标体系在全面、客观地描述数据库指纹综合性能的同时,清晰、准确地反映了指标间的逻辑关系。提出基于集成分类的数据库指纹综合评价算法,利用传统评价算法直接获取评价结果,并提出一种基于轮廓系数和K均值聚类的分级算法,解决不同评价算法间评价结果量纲不一致的问题。利用分级评价结果和分类算法构建集成分类器,实现对数据库指纹的精确分级综合评价。实验结果表明,基于集成分类的评价算法模型较同类分类器具有更好的准确度和泛化能力,能够有效地指导决策者选取性能更优的数据库指纹。Since database digital fingerprints with higher performance need to be selected for specific usage scenarios, and the theoretical system of database digital fingerprints need to be improved, it is of great practical significance to conduct comprehensive evaluation on database digital fingerprints. First, aiming at the usage scenarios, we use the performance evaluation indexes of traditional digital fingerprints to establish a hierarchical evaluation index system for database digital fingerprints, and propose a solution for index quantification and pre-processing of evaluation data. The index system can clearly and accurately reflect the logical relationships between the indicators, and comprehensively and objectively describe the performance of database digital fingerprints. Second, a comprehensive evaluation algorithm based on the integrated classification of database digital fingerprints is proposed to directly obtain evaluation results by the traditional evaluation algorithms, and a grading algorithm based on contour coefficients and K-mean clustering is proposed to solve the problem of inconsistent evaluation results between different evaluation algorithms. Third, the ensemble classifier is constructed by using the grading evaluation results and the classification algorithm to realize the accurate grading and comprehensive evaluation on database digital fingerprints. The experimental results show that the evaluation algorithm model based on ensemble classification has better accuracy and generalization than similar classifiers, and can effectively guide decision makers to select digital fingerprints of databases with better performance.

关 键 词：数字指纹 综合评价 集成分类 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]