Acquisition costs are so high in the Telecom business that firms are forced to focus on proper customer retention to keep up profitability. This paper seeks to answer which features best explain the reason behind customer churn and which machine learning algorithms best solve the task of data mining. In many journals different machine learning algorithms have been tested and shown significant results on identifying churn. This paper seeks to test some of the same models on new data to test their prediction rate on data that contains both demographic and expenses variables.
Through descriptive analysis it is shown that there might be some subtle indicators of positive relations between customer retention and usage of subscription. There is also evidence that a change in charges through discounts, can be a positive driver for customer retention.
It’s concluded that none of the chosen algorithms were able to predict the data precisely, since the best model only predicted 67.6% of the cases correctly. The reason for this is suspected to be due to poor data, though the results did hint that age of equipment and customer seniority are features that best explain churn tendencies at least in this case.
|Educations||MSc in Business Administration and Management Science, (Graduate Programme) Final Thesis|
|Number of pages||85|
|Supervisors||Niels Buus Lassen|