Bridging behavioural models and explainable AI in cryptocurrency adoption: a study of emerging markets with evidence from Vietnam

Abstract

Cryptocurrencies have become mainstream financial instruments, yet adoption remains uneven in emerging markets such as Vietnam, where rapid digitalization and regulatory uncertainty shape user behavior. Existing studies either use behavioral models that offer theoretical clarity but assume linear effects, or machine learning models that capture complexity but lack interpretability. This study combines behavioral theory with explainable artificial intelligence to examine cryptocurrency adoption in Vietnam using survey data from 1,039 respondents. Ten supervised learning algorithms were tested through repeated cross validation, and Random Forest delivered the highest accuracy. SHapley Additive exPlanations were used to interpret model outputs. Results show that trust, perceived usefulness, behavioral control, and financial literacy are key predictors, while perceived risk follows a curvilinear pattern. Interaction analysis reveals that usefulness rises with stronger behavioral control, and trust reduces risk only to a certain point. The study offers a theory informed and interpretable machine learning framework.Cryptocurrencies have become mainstream financial instruments, yet adoption remains uneven in emerging markets such as Vietnam, where rapid digitalization and regulatory uncertainty shape user behavior. Existing studies either use behavioral models that offer theoretical clarity but assume linear effects, or machine learning models that capture complexity but lack interpretability. This study combines behavioral theory with explainable artificial intelligence to examine cryptocurrency adoption in Vietnam using survey data from 1,039 respondents. Ten supervised learning algorithms were tested through repeated cross validation, and Random Forest delivered the highest accuracy. SHapley Additive exPlanations were used to interpret model outputs. Results show that trust, perceived usefulness, behavioral control, and financial literacy are key predictors, while perceived risk follows a curvilinear pattern. Interaction analysis reveals that usefulness rises with stronger behavioral control, and trust reduces risk only to a certain point. The study offers a theory informed and interpretable machine learning framework.