Credit card transactions, especially when linked to smart devices and the IoT ecosystem in general, are one of the drivers of contemporary digital economy as well as a major indicator of the overall financial activity. As such as well as for a plethora of other reasons it is imperative that fraudulent transactions be efficiently and reliably discovered. Because of their interconnected and time-dependent nature, a graphic representation not only is convenient, but also lends itself to machine learning strategies. To this end one viable approach is to construct a framework consisting of three steps. First, at each vertex a vector containing first and higher order attributes is embedded, then vertices are clustered, and finally vertex classification is done. As a concrete example three graph partitioning algorithms were selected, namely kNN, DBSCAN, and spectral clustering, whereas vertex clustering has been performed through logistic regression. The experimental results corroborate the efficiency of the abovementioned framework and are encouraging for the development of more higher order fraudulent transaction methods towards a more robust and highly reliable digital economy.