Koyon Injini na Taimako don Gano Tsarin da Ake Amfani da Shi don Kiyasin Ƙarfin Jiki na Ƙarshe (UTS) a cikin Samfuran PLA da Aka Bugawa ta Hanyar FDM

2.1. Kera Samfura & Sigogi

An samar da bayanai daga samfuran PLA 31 da aka kera ta hanyar FDM. An bambanta sigogi guda huɗu na tsari don ƙirƙirar saitin fasali don samfuran ML:

• Kashi na Ciki: Yawan gina ciki.
• Tsayin Layer: Kaurin kowane layer da aka ajiye.
• Gudun Buga: Gudun tafiyar bututu yayin ajiyewa.
• Zafin Fitarwa: Zafin filament ɗin da ya narke.

An auna UTS na kowane samfurin ta hanyar gwaji sannan aka rarraba su zuwa rukunoni (misali, "High" ko "Low" UTS) don ƙirƙirar matsalar rarrabuwa mai kulawa.

2.2. Hanyoyin Koyon Injini

An aiwatar da hanyoyin rarrabuwa masu kulawa guda huɗu daban-daban kuma aka kwatanta su:

1. Rarrabuwa na Logistic: Samfurin layi don rarrabuwa biyu.
2. Rarrabuwa na Haɓaka Gradient: Dabarar haɗaɗɗiya wacce ke gina bishiyoyi a jere don gyara kurakurai.
3. Bishiyar Yanke Shawara: Samfurin da ba na sigogi ba wanda ke raba bayanai bisa ƙimar fasali.
4. K-Nearest Neighbor (KNN): Algorithm na koyo na misali wanda ke rarraba ma'ana bisa ga mafi yawan ajin 'makwabta' k 'mafi kusa a cikin sararin fasali.

An kimanta aikin samfurin ta amfani da ma'auni kamar Maki F1 da Yankin Ƙarƙashin Lankwasa (AUC) na Halayen Aiki na Mai Karɓa (ROC).

3.1. Kwatanta Ayyukan Algorithm

Sakamakon gwaji ya ba da tsari bayyananne na ingancin samfuri don wannan aiki na musamman:

Yayin da Bishiyar Yanke Shawara ta yi daidai da makin F1 na KNN, ma'aunin AUC ya nuna mafi girman ikon KNN na bambanta tsakanin rukunonin UTS a duk matakan rarrabuwa.

3.2. Mafi Girman K-Nearest Neighbor

Algorithm ɗin KNN ya fito a matsayin samfurin da ya fi dacewa. Ana iya danganta nasararsa da yanayin bayanan da matsalar:

• Kama da Na Gida: UTS yana yiwuwa an ƙaddara shi ta hanyar rikitarwa, hulɗar da ba ta layi ba tsakanin sigogi. Kama na gida na KNN yana ɗaukar waɗannan tsare-tsaren ba tare da ɗaukar tsarin aiki na duniya ba, sabanin samfuran layi (Logistic Regression).
• Ƙarfi ga Ƙananan Bayanai: Tare da maki bayanai 31 kawai, samfuran da ba na sigogi ba masu sauƙi kamar KNN da Bishiyoyin Yanke Shawara ba su da sauƙin yin wuce gona da iri idan aka kwatanta da hadaddun hanyoyin haɗaɗɗiya kamar Haɓaka Gradient, waɗanda ƙila suna buƙatar ƙarin bayanai don yin gabaɗaya yadda ya kamata.
• Fahimta da Aiki: Yayin da Bishiyar Yanke Shawara ke ba da bayyanannen fassarar tushen ƙa'ida, aikinta (AUC) ya ɗan yi ƙasa da na KNN, yana nuna cewa tunanin KNN na tushen nisa ya fi dacewa da yanayin bayanan da ke ƙasa don wannan aikin hasashen kaddarorin.

Bayanin Chati (An nuna): Chati na sanduna zai yi tasiri sosai wajen ganin makin F1 (duk suna da 0.71 ga KNN da DT) kuma wani chati na sanduna ko tebur zai haskaka mai banbancewa: makin AUC, tare da sandar KNN ta fi sauran girma sosai (0.79), yana nuna bayyananne mafi girman ikon rarrabuwa.

4.1. Tsarin Lissafi

Za a iya tsara ainihin algorithm ɗin KNN don rarrabuwa. Idan aka ba da sabon vector na fasali na shigarwa $\mathbf{x}_{\text{new}}$ (wanda ya ƙunshi kashi na ciki %, tsayin layer, da sauransu), an ƙaddara ajinsa $C$ ta hanyar:

1. Lissafin Nisa: Lissafa nisa (misali, Euclidean) tsakanin $\mathbf{x}_{\text{new}}$ da duk vectors na horo $\mathbf{x}_i$ a cikin bayanan:

   $d_i = ||\mathbf{x}_{\text{new}} - \mathbf{x}_i||_2$

2. Gano Makwabta: Gano samfuran horo $k$ tare da mafi ƙanƙanta nisa $d_i$.
3. Mafi Yawan Zaɓe: Sanya ajin $C$ wanda ya fi yawa a cikin waɗannan makwabta $k$:

   $C(\mathbf{x}_{\text{new}}) = \arg\max_{c} \sum_{i=1}^{k} I(C_i = c)$

   inda $I(\cdot)$ shine aikin nuna alama, kuma $C_i$ shine ajin makwabcin $i$.

Ma'aunin AUC, inda KNN ya yi fice, yana wakiltar yuwuwar cewa samfurin yana sanya misali mai kyau na bazuwar sama fiye da misali mara kyau na bazuwar. AUC na 0.79 yana nuna damar 79% na daidaitaccen matsayi, yana nuna kyakkyawar ikon rarrabuwa.

4.2. Misalin Tsarin Bincike

Yanayi: Injiniya yana son hasashen ko sabon saitin sigogin FDM zai samar da "High" ko "Low" UTS ba tare da bugawa ba.

Aiwatar da Tsarin Aiki (Ba Code ba):

1. Wakilcin Bayanai: Sabon saitin sigogi {Ciki: 80%, Tsayin Layer: 0.2mm, Gudu: 60mm/s, Zafi: 210°C} an tsara shi azaman vector na fasali.
2. Tambayar Samfuri: An cusa wannan vector cikin samfurin KNN da aka horar ($k=5$, ta amfani da nisa na Euclidean, fasali masu daidaitawa).
3. Binciken Makwabta: Samfurin yana lissafin nisa zuwa duk bugu na tarihi 31. Ya samo bugu 5 mafi kama da na baya bisa kusancin sigogi.
4. Yanke Shawara & Amincewa: Idan 4 daga cikin waɗannan bugu 5 na baya suna da "High" UTS, samfurin yana hasashen "High" don sabon saitin. Rabon (4/5 = 80%) yana aiki azaman makin amincewa. Makin AUC na 0.79 yana ba da amincewa gabaɗaya ga ikon samfurin na sanya matsayi a duk yuwuwar bakin kofa.
5. Aiki: Injiniya yana amfani da wannan hasashen don amincewa da sigogin don wani sashi mai mahimmanci ko yanke shawarar daidaita su kafin bugu mai tsada.