{"id":1215,"date":"2023-07-27T06:34:51","date_gmt":"2023-07-27T06:34:51","guid":{"rendered":"https:\/\/statorials.org\/tr\/rde-cok-degiskenli-uyarlanabilir-regresyon-egrileri\/"},"modified":"2023-07-27T06:34:51","modified_gmt":"2023-07-27T06:34:51","slug":"rde-cok-degiskenli-uyarlanabilir-regresyon-egrileri","status":"publish","type":"post","link":"https:\/\/statorials.org\/tr\/rde-cok-degiskenli-uyarlanabilir-regresyon-egrileri\/","title":{"rendered":"R'de \u00e7ok de\u011fi\u015fkenli uyarlanabilir regresyon spline'lar\u0131"},"content":{"rendered":"

<\/p>\n

\n

\u00c7ok de\u011fi\u015fkenli uyarlamal\u0131 regresyon e\u011frileri<\/a> (MARS), bir dizi \u00f6ng\u00f6r\u00fcc\u00fc de\u011fi\u015fken ile bir yan\u0131t de\u011fi\u015fkeni<\/a> aras\u0131ndaki do\u011frusal olmayan ili\u015fkileri modellemek i\u00e7in kullan\u0131labilir.<\/span><\/p>\n

Bu y\u00f6ntem \u015fu \u015fekilde \u00e7al\u0131\u015f\u0131r:<\/span><\/p>\n

1.<\/strong> Bir veri k\u00fcmesini k<\/em> par\u00e7aya b\u00f6l\u00fcn.<\/span><\/p>\n

2.<\/strong> Her par\u00e7aya bir regresyon modeli yerle\u015ftirin.<\/span><\/p>\n

3.<\/strong> k<\/em> i\u00e7in bir de\u011fer se\u00e7mek \u00fczere k-katl\u0131 \u00e7apraz do\u011frulamay\u0131 kullan\u0131n.<\/span><\/p>\n

Bu e\u011fitimde, bir MARS modelinin R’deki bir veri k\u00fcmesine nas\u0131l s\u0131\u011fd\u0131r\u0131laca\u011f\u0131na ili\u015fkin ad\u0131m ad\u0131m bir \u00f6rnek sa\u011flanmaktad\u0131r.<\/span><\/p>\n

Ad\u0131m 1: Gerekli paketleri y\u00fckleyin<\/strong><\/h3>\n

Bu \u00f6rnek i\u00e7in ISLR \u00dccret<\/strong> veri k\u00fcmesini kullanaca\u011f\u0131z .  <\/em><\/strong> 3.000 ki\u015finin y\u0131ll\u0131k maa\u015f\u0131n\u0131n yan\u0131 s\u0131ra ya\u015f, e\u011fitim, \u0131rk gibi \u00e7e\u015fitli belirleyici de\u011fi\u015fkenleri i\u00e7eren paket.<\/span><\/p>\n

Verilere bir MARS modeli yerle\u015ftirmeden \u00f6nce gerekli paketleri y\u00fckleyece\u011fiz:<\/span><\/p>\n

 library<\/span> (ISLR) #contains Wage dataset<\/span>\nlibrary<\/span> (dplyr) #data wrangling<\/span>\nlibrary<\/span> (ggplot2) #plotting<\/span>\nlibrary<\/span> (earth) #fitting MARS models<\/span>\nlibrary<\/span> (caret) #tuning model parameters<\/span>\n<\/strong><\/pre>\n
 2. Ad\u0131m: Verileri g\u00f6r\u00fcnt\u00fcleyin<\/strong><\/h3>\n
 Daha sonra, \u00fczerinde \u00e7al\u0131\u015ft\u0131\u011f\u0131m\u0131z veri k\u00fcmesinin ilk alt\u0131 sat\u0131r\u0131n\u0131 g\u00f6r\u00fcnt\u00fcleyece\u011fiz:<\/span><\/p>\n
 #view first six rows of data<\/span>\nhead<\/span> (Wage)\n\n       year age maritl race education region\n231655 2006 18 1. Never Married 1. White 1. < HS Grad 2. Middle Atlantic\n86582 2004 24 1. Never Married 1. White 4. College Grad 2. Middle Atlantic\n161300 2003 45 2. Married 1. White 3. Some College 2. Middle Atlantic\n155159 2003 43 2. Married 3. Asian 4. College Grad 2. Middle Atlantic\n11443 2005 50 4. Divorced 1. White 2. HS Grad 2. Middle Atlantic\n376662 2008 54 2. Married 1. White 4. College Grad 2. Middle Atlantic\n             jobclass health health_ins logwage wage\n231655 1. Industrial 1. <=Good 2. No 4.318063 75.04315\n86582 2. Information 2. >=Very Good 2. No 4.255273 70.47602\n161300 1. Industrial 1. <=Good 1. Yes 4.875061 130.98218\n155159 2. Information 2. >=Very Good 1. Yes 5.041393 154.68529\n11443 2. Information 1. <=Good 1. Yes 4.318063 75.04315\n376662 2. Information 2. >=Very Good 1. Yes 4.845098 127.11574\n<\/strong><\/pre>\n
 3. Ad\u0131m: MARS modelini olu\u015fturun ve optimize edin<\/strong><\/span><\/h3>\n
 Daha sonra, bu veri k\u00fcmesi i\u00e7in MARS modelini olu\u015fturaca\u011f\u0131z ve hangi modelin en d\u00fc\u015f\u00fck test RMSE’sini (ortalama kare hatas\u0131) \u00fcretti\u011fini belirlemek i\u00e7in k-katl\u0131 \u00e7apraz do\u011frulama<\/a> ger\u00e7ekle\u015ftirece\u011fiz.<\/span><\/p>\n
 #create a tuning grid\n<\/span>hyper_grid <- expand. grid<\/span> (degree = 1:3,\n                          nprune = seq<\/span> (2, 50, length.out = 10) %>%<\/span>\nfloor<\/span> ())\n\n#make this example reproducible\n<\/span>set.seed(1)\n\n#fit MARS model using k-fold cross-validation\n<\/span>cv_mars <- train(\n  x = subset(Wage, select = -c(wage, logwage)),\n  y = Wage$wage,\n  method = \" earth<\/span> \",\n  metric = \" RMSE<\/span> \",\n  trControl = trainControl(method = \" cv<\/span> \", number = 10),\n  tuneGrid = hyper_grid)\n\n#display model with lowest test RMSE<\/span>\ncv_mars$results %>%<\/span>\n  filter<\/span> (nprune==cv_mars$bestTune$nprune, degree =cv_mars$bestTune$degree)    \ndegree nprune RMSE Rsquared MAE RMSESD RsquaredSD MAESD\t\t\n1 12 33.8164 0.3431804 22.97108 2.240394 0.03064269 1.4554\n<\/span><\/strong><\/pre>\n
 Sonu\u00e7lardan, en d\u00fc\u015f\u00fck test MSE’sini \u00fcreten modelin yaln\u0131zca birinci dereceden etkilere (yani etkile\u015fim terimleri olmayan) ve 12 terime sahip bir model oldu\u011funu g\u00f6rebiliriz. Bu model 33,8164’l\u00fck<\/strong> ortalama karek\u00f6k hata (RMSE) \u00fcretti.<\/span><\/p>\n
 Dereceye ve terim say\u0131s\u0131na g\u00f6re RMSE testini g\u00f6rselle\u015ftirmek i\u00e7in bir grafik de olu\u015fturabiliriz:<\/span> <\/p>\n
 #display test RMSE by terms and degree<\/span>\nggplot(cv_mars)\n<\/span><\/strong><\/pre>\n
\"R'deki<\/p>\n
 Uygulamada, bir MARS modelini a\u015fa\u011f\u0131daki gibi di\u011fer model t\u00fcrleriyle uyarlayaca\u011f\u0131z:<\/span><\/p>\n
 Not:<\/strong> MARS modelini belirtmek i\u00e7in method=\u201dearth\u201d kulland\u0131k. Bu y\u00f6nteme ili\u015fkin belgeleri burada<\/a> bulabilirsiniz.<\/span><\/em><\/p>\n