基于改进麻雀搜索算法的特征选择在入侵检测中的应用蔡昊,李军华,周成(南昌航空大学江西省图像处理与模式识别重点实验室,南昌330063)[摘要]机器学习在入侵检测中发挥着至关重要的作用,特征选择作为机器学习的关键预处理步骤,受到广大研究者的关注。针对麻雀搜索算法寻优能力强但易陷入局部最优的问题,本文对特征编码、位置更新等策略进行改进,提出一种多策略融合的二进制麻雀搜索算法,结合决策树分类器构造封装式特征选择算法,从高维特征空间中选择具有代表性的特征,以提高模型的预测能力并降低时间成本。基于NSL-KDD和UNSW-NB15数据集进行了性能评估,实验结果表明:与多种特征选择算法相比,利用该算法进行特征选择后的数据具有最佳二分类效果。[关键词]麻雀搜索算法;入侵检测;特征选择;特征编码;位置更新[中图分类号]TP393[文献标志码]Adoi:10.3969/j.issn.2096-8566.2023.02.009[文章编号]2096-8566(2023)02-0070-08ApplicationofFeatureSelectionBasedonImprovedSparrowSearchAlgorithminIntrusionDetectionCAIHao,LIJun-hua,ZHOUCheng(KeyLaboratoryofJiangxiProvinceforImageProcessingandPatternRecognition,NanchangHangkongUniversity,Nanchang330063,China)Abstract:Machinelearningplaysanimportantroleinintrusiondetection.Asakeypreprocessingstepofmachinelearning,featureselectionhasattractedtheattentionofmanyresearchers.Aimingatsolvingtheproblemthatsparrowsearchalgorithmhasstrongoptimizationabilitybuteasytofallintolocaloptimization,thisstudyimprovedthestrategiesoffeatureencodingandlocationupdate.Amultistrategymixedbinarysparrowsearchalgorithmwasproposed,whichcombinedadecisiontreeclassifiertoconstructwrappedfeatureselectionalgorithmandselectedrepresentativefeaturesfromhigh-dimensionalfeaturespacetoimprovethepredictionabilityofthemodelandreducethetimecost.TheperformanceofthisalgorithmwasevaluatedbasedonNSL-KDDandUNSW-NB15datasets.Theexperimentalresultsshowthatcomparedwithvariousfeatureselectionalgorithms,thedataafterfeatureselectionusingthealgorithmproposedinthisworkhasthebestbinaryclassificationeffect.Keywords:sparrowsearchalgorithm;intrusiondetection;featureselection;featureencoding;locationupdate引言随着互联网的高速发展...