Μέθοδοι μελέτης του ρυθμού απώλειας πελατών και της αξίας συνολικού χρόνου ζωής πελάτη
Techniques for studying churn and customer lifetime value
KeywordsΠροβλέψεις ; Ρυθμός απώλειας ; Συνολική αξία χρόνου ζωής πελάτη ; Λογιστική παλινδρόμηση ; Γραμμικά μοντέλα
The subject of the present MSc Dissertation is the prediction of churn’s rate and customer’s life time value. In recent years many companies have at their disposal large amount of data. Those data can be used in order to make predictions about the churn’s rate and customer’s life time value. However, traditional methods that have been used until today cannot exploit the large volume of data been available today. The aim of this thesis is firstly a theoretically description of the modern models machine learning and then apply these models on a real dataset. In particular, in the first chapter there is a mention in the importance of predicting churn rate, customer life time value and classic techniques used today. In the next three there is a description of classification , regression and dimensional reduction techniques. In the fifth chapter we construct variables to predict customer’s churn, life time value, select the most important ones and apply all the techniques described in previous sections. The comparison of customer’s churn rate was done using the AUC criterion and for the customer life time value the RMSE criterion. The logistic regression model produced the best results for predicting customer’s churn rate. The LightGBM model was better suited than the Random Forest technique that confirms its usefulness.