<\/p>\n
Ortalama mutlak y\u00fczde hatas\u0131 (MAPE),<\/strong> modellerin tahmin do\u011frulu\u011funu \u00f6l\u00e7mek i\u00e7in yayg\u0131n olarak kullan\u0131l\u0131r. A\u015fa\u011f\u0131daki \u015fekilde hesaplan\u0131r:<\/span><\/p>\n MAPE<\/strong> = (1\/n) * \u03a3(|ger\u00e7ek \u2013 tahmin| \/ |ger\u00e7ek|) * 100<\/span><\/p>\n Alt\u0131n:<\/span><\/p>\n MAPE, yorumlanmas\u0131 ve a\u00e7\u0131klanmas\u0131 kolay oldu\u011fundan yayg\u0131n olarak kullan\u0131lmaktad\u0131r. \u00d6rne\u011fin %11,5’lik bir MAPE de\u011feri, tahmin edilen de\u011fer ile ger\u00e7ek de\u011fer aras\u0131ndaki ortalama fark\u0131n %11,5 oldu\u011fu anlam\u0131na gelir.<\/span><\/p>\n MAPE de\u011feri ne kadar d\u00fc\u015f\u00fck olursa, model de\u011ferleri o kadar iyi tahmin edebilir. \u00d6rne\u011fin, MAPE’si %5 olan bir model, MAPE’si %10 olan bir modelden daha do\u011frudur.<\/span><\/p>\n MAPE’yi hesaplamak i\u00e7in yerle\u015fik bir Python i\u015flevi yoktur, ancak bunu yapmak i\u00e7in basit bir i\u015flev olu\u015fturabiliriz:<\/span><\/p>\n Daha sonra bu i\u015flevi iki tablo i\u00e7in MAPE’yi hesaplamak i\u00e7in kullanabiliriz: biri ger\u00e7ek veri de\u011ferlerini i\u00e7eren, di\u011feri ise tahmin edilen veri de\u011ferlerini i\u00e7eren.<\/span><\/p>\n Sonu\u00e7lardan bu model i\u00e7in ortalama mutlak y\u00fczde hatas\u0131n\u0131n %10,8009<\/strong> oldu\u011funu g\u00f6rebiliriz. Yani tahmin edilen de\u011fer ile ger\u00e7ekle\u015fen de\u011fer aras\u0131ndaki ortalama fark %10,8009’dur.<\/span><\/p>\n MAPE’nin hesaplanmas\u0131 ve yorumlanmas\u0131 kolay olmas\u0131na ra\u011fmen kullan\u0131m\u0131n\u0131n iki potansiyel dezavantaj\u0131 vard\u0131r:<\/span><\/p>\n 1.<\/strong> Mutlak y\u00fczde hatas\u0131n\u0131 hesaplama form\u00fcl\u00fc |ger\u00e7ek tahmin| oldu\u011fundan \/ |ger\u00e7ek| bu, ger\u00e7ek de\u011ferlerden herhangi birinin s\u0131f\u0131r olmas\u0131 durumunda MAPE’nin tan\u0131mlanmayaca\u011f\u0131 anlam\u0131na gelir.<\/span><\/p>\n 2.<\/strong> MAPE d\u00fc\u015f\u00fck hacimli verilerle kullan\u0131lmamal\u0131d\u0131r. \u00d6rne\u011fin, bir \u00fcr\u00fcn i\u00e7in ger\u00e7ek talep 2 ve tahmin 1 ise mutlak y\u00fczde hata de\u011feri |2-1| \/ |2| = %50, bu da tahminde yaln\u0131zca 1 birim sapma olsa bile tahmin hatas\u0131n\u0131n olduk\u00e7a y\u00fcksek g\u00f6r\u00fcnmesine neden olur.<\/span><\/p>\n","protected":false},"excerpt":{"rendered":" Ortalama mutlak y\u00fczde hatas\u0131 (MAPE), modellerin tahmin do\u011frulu\u011funu \u00f6l\u00e7mek i\u00e7in yayg\u0131n olarak kullan\u0131l\u0131r. A\u015fa\u011f\u0131daki \u015fekilde hesaplan\u0131r: MAPE = (1\/n) * \u03a3(|ger\u00e7ek \u2013 tahmin| \/ |ger\u00e7ek|) * 100 Alt\u0131n: \u03a3 \u2013 \u201ctoplam\u201d anlam\u0131na gelen bir sembol n \u2013 \u00f6rneklem b\u00fcy\u00fckl\u00fc\u011f\u00fc ger\u00e7ek \u2013 verilerin ger\u00e7ek de\u011feri tahmin \u2013 tahmin edilen verilerin de\u011feri MAPE, yorumlanmas\u0131 ve a\u00e7\u0131klanmas\u0131 kolay […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[11],"tags":[],"class_list":["post-835","post","type-post","status-publish","format-standard","hentry","category-rehber"],"yoast_head":"\n\n
Python’da MAPE nas\u0131l hesaplan\u0131r<\/strong><\/h3>\n
import numpy<\/span> as np<\/span>\n\ndef<\/span> mape( actual<\/span> , pred<\/span> ): \n actual, pred = np.array(actual), np.array(pred)\n return<\/span> np.mean(np.abs((actual - pred) \/ actual)) * 100\n<\/strong><\/pre>\n actual = [12, 13, 14, 15, 15,22, 27]\npred = [11, 13, 14, 14, 15, 16, 18]\n\nmap(actual, pred)\n\n10.8009\n<\/strong><\/pre>\n
MAPE kullan\u0131m\u0131na ili\u015fkin \u00f6nlemler<\/strong><\/span><\/h3>\n