Customer churn analysis with classification algorithms in telecommunication sector

The digital universe is filled with massive amounts of data generated by users around the world. Different analysis and storage methods are needed for data coming from various sources and whose volume is increasing rapidly, and at this point, the concept of 'Big Data' emerges. Companies or institutions are storing more and more data depending on their processes and trying to extract value by analyzing this data. Today's business world is increasingly competitive and understanding customer behavior provides a significant advantage in sustainable competition. The success of companies largely depends on how well they can analyze existing customer data and extract meaningful information. In this context, "Customer Churn Analysis" studies are carried out in order to predict the customer leaving tendency and to reveal the situations that cause this trend. The aim of this thesis is to create a model that predicts customers who tend to diverge in the telecommunications industry with big data analysis technology. In this context, Apache Spark's machine learning library (MLlib) and various classification methods were applied to the data set obtained from a telecommunication company. Firstly, preliminary preparations were made on the data, and then Logistic Regression, Decision Trees, Random Forest, Gradient Boosting Machines methods were applied to the data set. When the results of the study were examined, it was seen that Gradient Boosting Machines method gave the most successful results in churn prediction with 86,34% accuracy, 90,82% AUC and 63,45% F-Score rates. It has been observed that the most important factor causing customer churn is remaining time to the end of the contract.