A Comparative Model for Predicting Customer Churn using Supervised Machine Learning

Churn in customers is an important area of concern for a majority of telecommunications companies. The telecoms industry is of special interest since it suffers annual churn rates of up to 30%. Models have been developed to deal with this problem especially when it is found in such an industry that is the telecoms industry which relies on customers that are not contract based. Predictive models therefore become key to better understand customer churn churn. Handling this issue, in this study the author will implement the SEMMA approach to determine the model with the highest possible accuracy, then choose the best model based on percentage accuracy. This project develops a churn prediction model that can help businesses anticipate which customers are most likely to churn. To discover the key causes of customer turnover, it will employ machine learning techniques such as Random Forest Classifier, Decision Trees, Ada Boost Classifier, SGD Classifier, Logistic Regression, K Neighbors Classifier, Cat Boost Classifier and Gradient Boosting Classifier algorithms. The dataset is comprised of customer demographics, service received and the sum total of their charges from the respective company. It is a Kaggle data set with over 21 attribute obtained from more than 7 000 clients.