This study deals with the problem of customer churn, particularly in the telecommunications industry, which is one of the most important problems in the business’ world, because it directly affects the company’s revenues, and early detection of customer churn enables the company to take proactive measures to retain them. To solve the problem of customer churn we employed 4 different models of Neural Networks, with various numbers of hidden layers. Hence, because one of the proposed models contains only one hidden layer, it belongs to standard machine learning algorithms, whereas the remaining models are considered as Deep Learning Models. Data balancing problem presents challenges for classifiers. We noticed a slight imbalance in the dataset we used. To solve this problem, we applied SMOTE and SMOTE-ENN balancing techniques. The results showed that the greater the depth neural network model, the better the results, but the improvement was not noticeable, and the evaluation metrics values were very close.