Ukuqonda Ukwahlulwa Kwefestile Yeefestile KuMsayino Wokwahlulwa Kwamahlathi

Itheyibhile yoQhagamshelwano

Isishwankathelo kunye 1. Intshayelelo

2. Imvelaphi kunye nezandulela

2.1. Ihlathi eliSebenzayo lokuzahlula

2.2. Indlela yoMtyikityo

3. Utyikityo lweNdlela yokuzahlula kwehlathi

4. Iimvavanyo zamanani

4.1. IiParameters Uhlalutyo lobuthathaka

4.2. Izinto eziluncedo ze-(K-)SIF ngaphezulu kwe-FIF

4.3. Idatha yokwenyani yokuFumana iBenchmark

5. Ingxoxo kunye nesiphelo, iiNkcazelo zeMpembelelo, kunye neeReferensi

Isihlomelo

A. Ulwazi olongezelelweyo malunga noMsayino

B. K-SIF kunye SIF Algorithms

C. Iimvavanyo zamanani ezongezelelweyo

4.1. IiParameters Uhlalutyo lobuthathaka

Siphanda ukuziphatha kwe-K-SIF kunye ne-SIF malunga neeparamitha zazo ezimbini eziphambili: ubunzulu bomsayino k kunye nenani leefestile ezahlulahlulayo ω. Ngenxa yendawo, uvavanyo kubunzulu luhlehlisiwe kwiCandelo C.1 kwiSihlomelo.

Indima yoMtyikityo wefestile yokwahlulwa. Inani leefestile ezahlulahlulayo livumela ukutsalwa kolwazi kumathuba athile (akhethwe ngokungacwangciswanga) kwedatha esisiseko. Ngaloo ndlela, kwi-node yomthi ngamnye, ugxininiso luya kuba kwinxalenye ethile yedatha, efana nayo yonke i-curves yesampuli ngeenjongo zokuthelekisa. Le ndlela iqinisekisa ukuba uhlalutyo lwenziwa kumacandelo athelekisekayo eedatha, ukubonelela ngendlela ecwangcisiweyo yokuhlola nokuthelekisa amaxesha ahlukeneyo okanye iimpawu kwiisampulu zeesampulu.

Siphonononga indima yale parameter ngeeseti zedatha ezimbini ezahlukeneyo ezivelisa iindidi ezimbini zeemeko ezingaqhelekanga. Eyokuqala ithathela ingqalelo iziphazamiso ezibekwe zodwa kwisithuba esincinci, ngelixa eyesibini iqulethe ezizingisileyo kuwo wonke umsebenzi weparametrization. Ngale ndlela, siqwalasela impatho ye-K-SIF kunye ne-SIF ngokumalunga neentlobo ezahlukeneyo zokungaqhelekanga.

Iseti yedatha yokuqala yakhiwe ngolu hlobo lulandelayo. Silinganisa imisebenzi eyi-100 rhoqo. Emva koko sikhetha ngokungakhethiyo i-90% yala magophe kunye nengxolo ye-Gaussian kwi-sub-interval; kwi-10% eseleyo yeegophe, songeza ingxolo ye-Gaussian kwelinye i-sub-interval, eyahlukileyo kweyokuqala. Ngokuchaneke ngakumbi:

• I-90% yeegophe, ezithathwa njengesiqhelo, ziveliswa ngokwe

ngo-ε(t) ∼ N (0, 1), b ∼ U([0, 100]) kunye no-U emele ulwabiwo olufanayo.

• I-10% yeegophe, ezithathwa njengezingaqhelekanga, ziveliswa ngokwe

apho ε(t) ∼ N (0, 1) kunye b ∼ U([0, 100]).

Silinganisa ngokungaqhelekanga kwi-90% yeendlela kunye µ = 0, σ = 0.5, kwaye sizithathe njengedatha eqhelekileyo. Emva koko, i-10% eseleyo ifaniswe ne-drift µ = 0.2, ukutenxa okusemgangathweni σ = 0.4, kwaye ithathwa njengedatha engaqhelekanga. Sibala i-K-SIF ngamanani ahlukeneyo eefestile eziqhekezayo, ezihluka ukusuka kwi-1 ukuya kwi-10, kunye nenqanaba le-truncation elilingana no-2 kunye no-N = 1, 000 inani lemithi. Uvavanyo luphindwa ngamaxesha e-100, kwaye sixela i-AUC ephakathi phantsi kwee-curves ze-ROC kuMfanekiso 1 kuzo zombini iiseti zedatha kunye nezichazi-magama ezithathu ezikhethwe kwangaphambili.

Kwiseti yedatha yokuqala, apho i-anomalies ibonakalisa kwinxalenye encinci yemisebenzi, ukwandisa inani lokwahlula kuphucula kakhulu ukusebenza kwe-algorithm ekuboneni ukungahambi kakuhle. Uphuculo lokusebenza lubonisa ithafa emva kweefestile ezisithoba eziqhekezayo. Kwimeko yedatha yesibini ene-anomalies eqhubekayo, inani eliphezulu leefestile eziqhekezayo linempembelelo esecaleni ekusebenzeni kwe-algorithm, igcina iziphumo ezanelisayo. Ngoko ke, ngaphandle kolwazi lwangaphambili malunga nedatha, ukukhetha inani eliphezulu leefestile eziqhekezayo, ezifana ne-10, ziyakuqinisekisa ukusebenza okuqinileyo kuzo zombini iimeko. Ukongeza, inani elibaluleke ngakumbi leefestile ezicazululiweyo lenza ukubalwa kotyikityo kwinxalenye encinci yemisebenzi, ekhokelela ekuphuculweni kobuchule bokubala.