Functional classification of bitcoin addresses

A classification model for predicting the main activity of bitcoin addresses based on their balances is proposed. Since are functions time, methods from functional data analysis applied; more specifically, features proposed principal components data. Classifying a relevant problem two reasons: to understand composition market, and identify used illicit activities. Although other classifiers have been proposed, they focus primarily network rather than curve behavior. The approach, hand, does not require any information prediction. Furthermore, advantage being straightforward build, unlike expert-built features. Results show improvement when combining with scalar features, similar accuracy models using those separately, which points good alternative domain-specific knowledge available.