Predicting Digital Transactions in Thailand Using Satellite Data and Machine Learning Methods

Vollständiger Abstract

Worum geht es in dieser Arbeit?

The digital economy in Thailand has experienced significant growth over the past decade. With the adoption of financial technology, monitoring this rapid expansion poses challenges for policymakers, particularly in predicting the trajectory of such growth given data compilation constraints. This paper aims to predict digital transactions in Thailand by utilizing alternative economic data, such as satellite imagery, and applying machine learning approaches to support more proactive policymaking strategies. Monthly data from 2018 to 2024, sourced from the Bank of Thailand and NASA satellite imagery, is employed. Key variables include internet banking, mobile banking, PromptPay usage, population density, GDP per capita, nighttime light intensity, daytime surface temperature, and nighttime surface temperature. The study compares the predictive performance of various machine learning algorithms, including artificial neural networks, Random Forests, and support vector machines. The dataset is split into training and testing subsets at a 70:30 ratio for validating the prediction. The results highlight the potential of satellite data in prediction, particularly the significant influence of nighttime light intensity and daytime surface temperature on digital transactions. Additionally, the Random Forest outperforms other algorithms due to its ability to capture complex associations, even in the presence of non-linear and irregular patterns.