Track: Data Analytics and Big Data

Abstract

Online educational service is a highly competitive service sector in Indonesia, with practice tests for university entrance examinations as its main product. In 2020, an online educational service firm is surveyed to have a high percentage of churned customers, at an average of 73%. Churned customer is defined as a customer who no longer uses a service or product, while in comparison, a benchmark churn rate for the whole education industry is merely 9.6%. It indicates a mismatch between expectation and actual service received by the customer. This research develops and implements a prediction model for churned customers using Cross Industry Standard Process for Data Mining (CRISP-DM) methodology. The selected model is built based on the Random Forest algorithm. Furthermore, a sentiment analysis model is also built, based on the Bidirectional Encoder Representations from Transformers (BERT) algorithm. The sentiment analysis model is trained and validated by text data from Twitter, which indicates customers’ perspectives on the practice test for university entrance exams. The final models show a 97% rate of accuracy and 100% for recall rate for churned customer prediction model and a 88% rate of accuracy for the sentiment analysis model. Lastly, a dashboard is developed and deployed to monitor online sentiments and their corresponding churn probability.