Predikce odlivu zákazníků pomocí metod strojového učení: případová studie na datech Netflix

This bachelor’s thesis focuses on the problem of customer churn prediction in the context of a digital subscription-based streaming service. The main objective of the thesis is to train, compare, and evaluate selected machine learning classification models and to assess their ability to identify customers with an increased risk of churn. The practical part of the thesis is based on a synthetic Netflix Customer Churn dataset, which simulates the behavior of users of a video-streaming platform. The workflow begins with data analysis and preprocessing, including data cleaning, handling missing values, feature transformations, and addressing potential class imbalance. Subsequently, several machine 4 learning classification methods are applied, namely logistic regression, decision tree, random forest, and gradient boosting. The models are evaluated using standard binary classification metrics. The results show that ensemble methods outperform simpler linear approaches in terms of predictive performance. Gradient Boosting is identified as the most suitable model, achieving the best balance between discriminatory power and performance stability. The analysis further indicates that user activity-related features play an important role in churn prediction. However, these findings should be interpreted with caution due to the synthetic nature of the dataset. The thesis demonstrates that machine learning models can serve as a valuable decision-support tool for customer retention strategies in digital subscription services.