加权Soft Voting多模型集成钓鱼网站检测模型

打开文本图片集

Abstract：Aimingattheproblemofinsuficientgeneralizationabilityofsinglemodelsinphishingwebsitedetectionthispaper proposesaweightedSoftVotingmulti-modelfusiondetectionmethodoptimizedbySLSQPalgorithm.Byintegratingsix heterogeneousbasemodels—XGBost，LightGBM，CatBoostRandomForest，GradientBoosting，andMLPClasifierthisethod usestheSLSQPalgorithmtooptimizetheweightsofeachmodelonthevalidationsetwiththegoalofmaximizing theAUC metric，constructinganintegrateddetectionsystemwithbothighdetectionrateandlowfalsepositiverate.Experimentalresults demonstratethatthefusionmodeloutperformssinglemodelsinaccuracy，recall，andF1-score，withanaccuracyof 95.22% and an AUCvalueof0.9762underthestatic featureset;afterintroducingdynamicextended features，theaccuracyisimproved to 96.75% （2withanAUCvalueof0.9845.Thismethodsignificantlyimprovestherobustnessanddetectionperformanceofphishingwebsite identification，andprovidesaneficient solutionforphishingattack defenseincomplex networkenvironments.

Keywords：PhishingWebsiteDetection;WeightedSoftVoting;Multi-ModelFusion;EnsembleLearning;SLSQPAlgorithm

0引言

随着互联网技术的快速发展和在线服务的普及，钓鱼网站已成为网络攻击的主要载体之一。（剩余5333字）