Developing a model for discovering the causes of customer churn from banking services via hybrid approach of data mining and survey

Customer churn management consist of three main phases: identifying churners, discovering the causes of churn and adapting appropriate strategies against this problem. Most of studies in this field focused on prediction of customer churn. Few studies about discovering causes of churn are just about testing primary hypothesis about probable causes. This study because of the shortage of previous studies in this field has made lots of innovations. Some of these innovations are: designing a new framework for discovering causes of churn and designing a hybrid approach from data mining and survey techniques which carried out without benchmarking from any similar study. Proposed framework includes four main steps: feature construction and selection, identification of churners, discovering the causes of churn, and validating the results. Current account customers of Keshavarzi Bank are selected as casestudy of this research and the required data is gathered trough questionnaire. The approach used for discovering causes of churn is extracting the rules which lead to churn in various clusters of customers. For this purpose, decision tree technique with target variable of churn label is utilized. Validating the results is carried out by testing it on validation data set and calculating the top lift and overall error rate. The extracted rules represent that there is a tendency to churn among big segment of keshavarzi's customers. The most important reason (specially among higher salary customers) is about manner of bank agents not reasons expressed by experts such as the way of lending or the profit of backup accounts.