摘要
针对在不平衡的贫困学生数据中,传统的机器算法在分类时会偏向于多数类而忽略少数类,而导致少数特困类的分类准确率低的问题,提出了一种基于代价敏感的贫困生分类方法(CMPSC)。首先采取基于特征选择的预处理方法对低质量的贫困生数据进行处理;然后使用基于遗传算法的搜索方法确定贫困生数据的最优代价敏感矩阵;最后由经预处理的贫困生数据构建兼顾少数贫困类的代价敏感分类器,降低多数贫困类的影响。使用多个真实广西贫困生数据集进行对比实验,以CART算法为基准线,CMPSC方法平均总体分类准确率浮动值为0.66%,平均特困类分类准确率提升率为6.3%,最高提升率可达14.7%。本文方法可以在保持总体分类准确率的同时,有效提高少数特困类的分类准确率。
From imbalanced data on students in poverty,the traditional machine learning algorithm is biased towards the majority classes in classifying,which leads to inaccuracy of high-needs student classes.A classification method was tailored for impoverished students based on cost-sensitivity.First,a feature selection-based preprocessing method was employed to address the low quality data.Then,the optimal cost-sensitive matrix was determined by the search method based on genetic algorithm.Finally,the preprocessing data were used to train a cost-sensitive classifier for minority classes with reduced impact of the majority class.Multiple datasets of actual impoverished students in Guangxi was wsed to conduct comparative experiments.With the CART algorithm as the baseline,the average overall accuracy of this method shoas a floating value of 0.66%.The average accuracy improvement of minority classes is 6.3%,and the highest improvement can reach 14.7%.The results show that the proposed method can improve the accuracy of student classification with accuracy.
作者
黄海艳
韦必忠
戴戬
肖子涵
HUANG Hai-yan;WEI Bi-zhong;DAI Jian;XIAO Zi-han(College of Computer Science and Information Security,Guilin University of Electronic Technology,Guilin 541004,China;Guangxi Cooperative Innovation Center of Cloud Computing and Big Data,Guilin University of Electronic Technology,Guilin 541004,China)
出处
《桂林理工大学学报》
CAS
北大核心
2022年第4期988-995,共8页
Journal of Guilin University of Technology
基金
国家自然科学基金项目(61861013)
广西创新驱动发展专项科技重大专项基金项目(桂科AA18118031)。
关键词
贫困生分类
代价敏感
遗传算法
不平衡数据
精准扶贫
impoverished student classification
cost sensitive
genetic algorithm
imbalanced dataset
targeted poverty alleviation