Abstract:
Churn Prediction uses data mining techniques to facilitate companies in consumer
retention. The purpose of this thesis research is to study prediction classifiers used in
data mining and apply them in case of ride-hailing industry to form Driver Churn
Prediction Model. Subscription-based companies in sectors like banking, media uses
it to know the customer behaviour so that they can serve them better and prevent the
churn. Companies spent a lot of time and money to acquire the customer and they need
to spend again on marketing to get a customer back or to get a new one. All companies
are interested that their customer should not leave and remain loyal to them. To achieve
this loyalty company needs to work on data and predict the churn and compensate them.
In this research study I have developed churn prediction model based on previously
available data. Transportation Network Providers dataset is used to train a Driver
Churn Prediction Model to predict driver churn. For prediction techniques such as
logistic regression, decision tree, random forest, Bayesian optimization and Neural
networks are fused together to generate a model to derive a model for churn prediction.
The accuracy score and Confusion matrix of each classifier is compared to show better
performing model. The outcomes gathered in this study also show the significance of
class imbalance problem, which is solved using SMOTE, in case of churn prediction.
