Churn prediction in English language learning application based on learner interaction

Mobile phones and apps have changed the landscape of e-learning and have revolutionized the way people learn a second language by facilitating anytime-anywhere learning, game-based resources and socially interactive learning activities. Despite these features and affordances, these language learning apps suffer a fate of high churn rates. In this project, we examined the churning behaviour of learners in the context of a language learning app called Hello English. We applied descriptive analytics to analyse the behavioural differences between churners and non-churners and studied their interaction with the app to early-predict churning behaviour. Our findings indicate that non-churners interact with the mobile app more frequently compared to churners. Also, the trained machine learning classifiers can predict learner churning behaviour with a high recall value (0.824) and F1 (0.778). This churn detection will enable the app developers to provide intervention for learner retention.

To summarize we have predicted churn in English language learning application based on learner interaction, specifically: 

• Performed quantitative analysis using trace data of a mobile language learning application to identify the difference between churners and non-churner, and male and female learners

• Applied machine learning model to predict learners churning behaviour

• Conducted a sensitivity analysis to early predict learner’s churn behaviour