Creative Commons

This dissertation was written as a part of the MSc in Data Science at the International Hellenic University in Thessaloniki, Greece . The goal of this thesis was to inform everyone about the challenge of churning in the telecommunication indus tries and underline how important it is for them to deal with it, since their revenue as well as their reputation are directly affected. The first section introduce s everyone to this global phenomenon and explains important facts regarding the process of j oining a telecom industry, how the company handles its clientele and finally, under what criteria is someone assumed to be a churner. The second pa rt consists of the literature review, where previous researchers analyze their insights, their selected class ifiers that were used to tackle this issue as well as the criteria to evaluate those classifiers. The third section is devoted to the exploratory data analysis of an online – acquired telecom dataset , where bar charts and plots are applied to discover patt erns, useful information that may explain this phenomenon as well as similarities with the already provided literature. The fourth part is more technical, since it introduces the machine learning part of this thesis, which will try to predict people that a re possible to churn. The modules as well as the parameters optimization are also stated in this section. Finally, the last part is composed of the conclusion, where a small revision of what it was previously said is stated as well as the limitations and a ny further improvement that can be done to find a more suitable solution to this phenomenon.