面向电子产品报废预测的特征选择方法  

Feature selection method of electronic product obsolescence forecasting

作　　者：赵敏 刘妍[1] ZHAO Min;LIU Yan(College of Computer Science and Technology,Changchun University of Science and Technology,Changchun Jilin 130022,China)

机构地区：[1]长春理工大学计算机科学技术学院,长春130022

出　　处：《计算机应用》2020年第S02期19-23,共5页journal of Computer Applications

基　　金：吉林省自然科学基金资助项目(20180101035JC)。

摘　　要：针对电子产品在进行报废预测的过程中出现的预测结果准确性低的问题,提出了一种用于电子产品报废预测的特征选择方法。首先,对电子产品进行报废模式和影响分析,根据电子产品潜在的报废模式以及每种报废模式可能的影响分析得到电子产品的相关特征;然后,采用选择消去方法(ELECTRE)来评价电子产品特征,在评价过程中,依据每个特征的严重程度、更新频率以及可预测程度构造弱支配关系来淘汰劣特征;最后,根据每个特征之间的优劣关系得到可用于报废预测的关键特征,以手机为例进一步解释了所提出的特征选择方法在实际中的应用。实验结果表明,与信息增益以及ReliefF特征选择算法相比,所提方法的平均预测准确性比信息增益高1.4个百分点,比ReliefF特征选择算法高6.2个百分点。For the problem of the low accuracy of the prediction results in the electronicprocess of product obsolescence forecasting,a feature selection method for obsolescence forecasting of electronic products was proposed.Firstly,the obsolescence mode and effects analysis was carried out for electronic products.According to the potential obsolescence mode of electronic products and the possible impact of each obsolescence mode,relevant characteristics of electronic products could be obtained.Then,the ELimination Et Choice Translating REality(ELECTRE)was used to evaluate the characteristics of electronic products.In the evaluation process,weak dominance relations were constructed based on the severity,update frequency and predictability of each feature to eliminate inferior features.Finally,according to the pros and cons of each feature,the key features that could be used for obsolescence forecasting were obtained.Mobile phone was taken as an example to further explain the application of the proposed feature selection method in practice.The experimental results show that the average prediction accuracy obtained by the proposed method is 1.4 percentage points higher than that of information gain and 6.2 percentage points higher than that of ReliefF feature selection algorithm.

关 键 词：特征选择 报废模式 选择消去法 报废预测 关键特征 电子产品 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]