A Classification Analysis-based Approach to Internet Adoption in Nepal

This paper examines the possible factors influencing internet adoption in Nepal, using the DHS USAID 2022 dataset. Three classifier models were applied to predict internet usage behavior: mobile financial transactions in the last 12 months, frequency of internet use, and general internet usage. Balancing techniques like SMOTE and evaluation methods such as XGBoost, permutation tests, and partial dependence analysis were employed. Our analysis reveals that geographical and technological factors consistently play the most significant roles in predicting mobile and internet usage behaviors, while gender and ethnic factors have a weaker impact on usage. Moreover, our results explicate that the relationship between digital inequality and socioeconomic factors is extremely nonlinear, and far from trivial. The nonlinear patterns observed in our partial dependence analysis reveal that digital adoption in the highest-educated group of people or those living in metropolitan areas, is disproportionately higher than other segments of people.