Makine Öğrenmesi ile Churn Talebi Analizi
No Thumbnail Available
Date
2025
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Open Access Color
OpenAIRE Downloads
OpenAIRE Views
Abstract
Son günlerde Pay TV hizmetleri çeşitlenmektedir. Çeşitlenen hizmetlerle birlikte şirketler arasındaki rekabette artmaktadır. Strait Reserch araştırmalarına göre 2021 yılında 183 milyon dolar olan Global Pay TV pazarının 2030 yılında 210 milyon dolara yükselmesi tahmin edilmiştir. Bunun yanısıra Neflix'in araştırmasına göre edinilmiş veya elde tutulan abonelerin daha değerli olduğu belirtilmektedir. Aboneliklerini sonlandırmayı düşünen müşterilerin önceden tahmin edilebilmesi şirketler açısından oldukça önemli bir hal almıştır. Son yıllarda etkinliği gittikçe artan makine öğrenmesi ve yapay sinir ağları gibi tekniklerle tahminlemeyi yapmak daha kolay hale gelmiş ve şirketlere önemli katkılar sağlamaktadır. Bu çalışmada, bazı makine öğrenmesi modelleri ile ilgili bilgiler verilmiştir. Pay TV hizmeti veren bir şirketin verileri kullanılarak bazı makine öğrenmesi modelleri ile tahminlemeler yapılmıştır. Son olarak abonelik iptal talep edebilecek müşterilerin, son yaptığı abonelik iptal talebi ve en çok yaptığı iptal talepleri ile ilgili bilgilerde sonuca eklenerek, abonelik iptal talebi gerçekleşmeden, müşteriyle iletişime geçilmesi ve uygun bir kampanya önerilerek iptali önlemek hedeflenmiştir. Bu çalışma da etiketli veriler kullanılarak denetimli öğrenme teknikleri ile çalışılmıştır. Çalışma kapsamında Random Forest, XGBoost Classifier, KNeigbors Classifier, Logistic Regression, Ada Boost Classifier, Linear Discriminant Analysis, Decision Tree Classifier ve Extra Tree Classifier modelleri kullanılmıştır. Bütün modeller için başarı ölçütleri bulunmuş ve karşılaştırılmıştır. Servis sağlayıcıdan elde edilen veriler açısından en uygun model Random Forest olarak belirlenmiştir.
Recently, Pay TV services are diversifying. With diversified services, competition between companies increases. According to a study, the size of Pay TV is $183 million in 2021. According to the same study, this is estimated to reach $210 million in 2030. [1] In addition, according to Netflix's research, it is stated that acquired or retained subscribers are more valuable. [2] Predicting customers who are considering terminating their subscriptions has become very important for companies. In recent years, ML and artificial neural network techniques have made significant progress in prediction. They provide significant contributions to companies' decision-making processes. In this study, live data from a company that provides services in Turkey was used, and predictions were made with 8 different machine learning. Finally, by adding information about the last subscription cancellation request and the most frequently made cancellation requests of customers who may request subscription cancellation, the aim is to contact the customer before the subscription cancellation request is realized and to prevent cancellation by offering a suitable campaign. In this study, supervised learning techniques were used using unlabeled data. Within the scope of the study, Random Forest, XGBoost Classifier, KNeighbors Classifier, Logistic Regression, Ada Boost Classifier, Decision Tree Classifier, Extra Tree Classifier and Linear Discriminant Analysis models were used. Success criteria were found and compared for all models. When the data used in the study and the results of 8 different models were compared, the Extra Tree model was selected from among the few models that gave the best results. The purpose of choosing this model is to obtain better results than other models for records with fewer samples.
Recently, Pay TV services are diversifying. With diversified services, competition between companies increases. According to a study, the size of Pay TV is $183 million in 2021. According to the same study, this is estimated to reach $210 million in 2030. [1] In addition, according to Netflix's research, it is stated that acquired or retained subscribers are more valuable. [2] Predicting customers who are considering terminating their subscriptions has become very important for companies. In recent years, ML and artificial neural network techniques have made significant progress in prediction. They provide significant contributions to companies' decision-making processes. In this study, live data from a company that provides services in Turkey was used, and predictions were made with 8 different machine learning. Finally, by adding information about the last subscription cancellation request and the most frequently made cancellation requests of customers who may request subscription cancellation, the aim is to contact the customer before the subscription cancellation request is realized and to prevent cancellation by offering a suitable campaign. In this study, supervised learning techniques were used using unlabeled data. Within the scope of the study, Random Forest, XGBoost Classifier, KNeighbors Classifier, Logistic Regression, Ada Boost Classifier, Decision Tree Classifier, Extra Tree Classifier and Linear Discriminant Analysis models were used. Success criteria were found and compared for all models. When the data used in the study and the results of 8 different models were compared, the Extra Tree model was selected from among the few models that gave the best results. The purpose of choosing this model is to obtain better results than other models for records with fewer samples.
Description
Keywords
Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve Kontrol, Kayıp Müşteri Tahminlemesi, Tahminleme, Yapay Zeka ve Makine Öğrenmesi Dersi, Computer Engineering and Computer Science and Control, Customer Churn Prediction, Forecasting, Artificial Intelligence and Machine Learning Course
Turkish CoHE Thesis Center URL
Fields of Science
Citation
WoS Q
Scopus Q
Source
Volume
Issue
Start Page
End Page
66
Collections
Google Scholar™
Sustainable Development Goals
3
GOOD HEALTH AND WELL-BEING

4
QUALITY EDUCATION

6
CLEAN WATER AND SANITATION

16
PEACE, JUSTICE AND STRONG INSTITUTIONS
