Decoding Gutandukanya Idirishya Sensitivity mumashyamba yo kwigunga

Imbonerahamwe

Ibisobanuro na 1. Intangiriro

2. Amavu n'amavuko

2.1. Ishyamba ryimikorere

2.2. Uburyo bwo gusinya

3. Umukono wo Kwigunga Amashyamba Uburyo

4. Ubushakashatsi Bwinshi

4.1. Ibipimo Isesengura Rikomeye

4.2. Ibyiza bya (K-) SIF hejuru ya FIF

4.3. Ibipimo-byukuri Anomaly Kumenyekanisha Ibipimo

5. Ikiganiro & Umwanzuro, Amatangazo Yingaruka, na Reba

Umugereka

A. Amakuru yinyongera kubyerekeye umukono

B. K-SIF na SIF Algorithms

C. Ubushakashatsi Bwiyongereye

4.1. Ibipimo Isesengura Rikomeye

Dutohoza imyitwarire ya K-SIF na SIF kubijyanye nibintu bibiri byingenzi: ubujyakuzimu bwumukono k numubare wamadirishya yatandukanijwe ω. Kubwikibanza, igeragezwa ryimbitse ryimuriwe mu gice C.1 kumugereka.

Uruhare rwumukono Gutandukanya Idirishya. Umubare wamadirishya yatandukanijwe yemerera gukuramo amakuru hejuru yintera yihariye (yatoranijwe kubushake) yamakuru yibanze. Rero, kuri buri giti node, intumbero izaba kumurongo runaka wamakuru, ibyo bikaba bimwe murugero rwicyitegererezo cyose kugirango ugereranye. Ubu buryo buteganya ko isesengura rikorwa ku bice byagereranijwe byamakuru, bigatanga inzira ihamye yo gusuzuma no kugereranya intera zitandukanye cyangwa ibiranga hirya no hino.

Turasesengura uruhare rwibi bintu hamwe na dataseti ebyiri zitandukanye zerekana ubwoko bubiri bwibintu bidasanzwe. Iya mbere ireba ibintu bidasanzwe mu ntera ntoya, mugihe icya kabiri kirimo ibintu bikomeza kumikorere yose yibikorwa. Muri ubu buryo, twitegereza imyitwarire ya K-SIF na SIF kubijyanye nubwoko butandukanye bwa anomalies.

Dataset yambere yubatswe kuburyo bukurikira. Twigana ibikorwa 100 bihoraho. Noneho duhitamo kubushake 90% yiyi mirongo hamwe n urusaku rwa Gaussia kurwego ruto; kuri 10% isigaye yumurongo, twongeyeho urusaku rwa Gaussiya kurindi sub-intera, itandukanye niyambere. Byukuri:

• 90% y'imirongo, ifatwa nkibisanzwe, ikorwa ukurikije

hamwe na ε (t) ∼ N (0, 1), b ∼ U ([0, 100]) na U byerekana isaranganya rimwe.

• 10% y'imirongo, ifatwa nkibidasanzwe, ikorwa ukurikije

aho ε (t) ∼ N (0, 1) na b ∼ U ([0, 100]).

Twigana kuri 90% yinzira hamwe na µ = 0, σ = 0.5, kandi tubifata nkamakuru asanzwe. Hanyuma, 10% isigaye bigereranwa na drift µ = 0.2, gutandukana bisanzwe σ = 0.4, kandi bifatwa nkamakuru adasanzwe. Turabara K-SIF hamwe numubare utandukanye wamadirishya yatandukanijwe, bitandukanye kuva 1 kugeza 10, hamwe nurwego rwo kugabanya rwashyizweho rungana na 2 na N = 1, 000 umubare wibiti. Ubushakashatsi busubirwamo inshuro 100, kandi turatanga raporo ya AUC yagereranijwe munsi yumurongo wa ROC ku gishushanyo 1 kuri datasets zombi hamwe ninkoranyamagambo eshatu zabanjirijwe.

Kuri dataset yambere, aho anomalies igaragara mugice gito cyimirimo, kongera umubare wamacakubiri byongera cyane imikorere ya algorithm mugutahura ibintu bidasanzwe. Iterambere ryimikorere ryerekana ikibaya nyuma yicyenda icyenda. Kubijyanye na dataset ya kabiri hamwe na anomalies idahwema, umubare munini wamadirishya yatandukanijwe agira ingaruka nke kumikorere ya algorithm, bikomeza ibisubizo bishimishije. Kubwibyo, udafite ubumenyi bwambere kubyerekeranye namakuru, uhitamo umubare munini ugereranije wamadirishya yatandukanijwe, nka 10, byemeza imikorere ikomeye muribintu byombi. Byongeye kandi, umubare munini cyane wamacakubiri ya Windows atuma kubara umukono ku gice gito cyimirimo, biganisha ku kubara neza.