Prediction of Customer Engagement Response to E-wallet Content Based on Machine Learning Using Combined E-Wallet Dataset and Individual E-Wallet Dataset

Abstract

Objectives: The COVID pandemic proved that humans are not hesitant to adopt technology in the banking industry, especially as a method of payment. In Indonesia, this fact is demonstrated by the value of e-wallet transactions in 2021 which reached 18,5 billion USD. The implementation of QRIS (Quick Response Code Indonesian Standard) in the payment systems of e-wallets compels every e-wallet provider to intensify their marketing activities to overcome competitors. Previous studies on this topic primarily focused on factors that influence customer engagement on social media using traditional approaches, such as interviews and statistical methods to process the data. This research aims to overcome the limitations of previous research by using machine learning methods in predicting the customer engagement type of primary data collected directly from social media in this case Twitter.Methodology: In this paper, we propose the application of machine learning methods such as XGBoost, Random Forest, Decision Tree, and KNN to predict the most likely engagement type of a tweet related to e-wallet content using 15.756 data which are directly collected from Twitter.Finding: This research successfully found that XGBoost and KNN are the machine learning algorithms that perform best and the results in prediction from using the combined dataset and the individual e-wallet brands dataset are similar.Conclusion: Even though the prediction accuracy in this research is good, this research still has many limitations. Thus, future research in the same field would benefit from a larger amount of data to accommodate machine learning algorithms that are more complex like deep learning.