Making applications aware of the mobility experienced by the user can open\nthe door to a wide range of novel services in different use-cases, from smart\nparking to vehicular traffic monitoring. In the literature, there are many\ndifferent studies demonstrating the theoretical possibility of performing\nTransportation Mode Detection (TMD) by mining smart-phones embedded sensors\ndata. However, very few of them provide details on the benchmarking process and\non how to implement the detection process in practice. In this study, we\nprovide guidelines and fundamental results that can be useful for both\nresearcher and practitioners aiming at implementing a working TMD system. These\nguidelines consist of three main contributions. First, we detail the\nconstruction of a training dataset, gathered by heterogeneous users and\nincluding five different transportation modes; the dataset is made available to\nthe research community as reference benchmark. Second, we provide an in-depth\nanalysis of the sensor-relevance for the case of Dual TDM, which is required by\nmost of mobility-aware applications. Third, we investigate the possibility to\nperform TMD of unknown users/instances not present in the training set and we\ncompare with state-of-the-art Android APIs for activity recognition.\n