This talk will showcase the major outputs of the research project “Safari Njema: Supporting Sub-Saharan paratransit mobility policies through big data analysis”. The project is a pilot interdisciplinary project that took place in 2019-2021 in Greater Maputo (Mozambique). It proposes an innovative methodology to study and reduce “transport poverty”, supporting bottom-up and ordinary practices leveraging on new data sources like GPS traces by mobile phones and open crowd-sourced data that could possibly replace traditional data sources (like census or survey data) which are often absent or at least not reliable in this geographical and social context. The project was coordinated by a highly interdisciplinary team based at the Politecnico di Milano, composed by scholars and researchers in the field of mathematics and statistics, urban studies, policy design and co-design, information engineering and management engineering. The project lists also a wide group of diversified partners which includes other academic institutions (Universidade Eduardo Mondlane and University College London), local and international public institutions (Metropolitan Transport Agency of Maputo, UN-habitat, Italian Agency for Development and Cooperation, Wasa Think-and-do Tank), and private international companies (Cuebiq and Vodacom Mozambique). The talk will be divided into two parts. In the former one we will give a quick introduction to the applicative context and present problems and opportunities associated to the informal collective and self-organized mobility system (i.e., paratransit mobility) which 80% of the Grater Maputo population relies on. We will then describe the objectives of the Safari Njema project and describe some of the outputs of the project derived from the analysis and modelling of the data mentioned in the previous paragraph. The latter part of the talk will focus instead on a specific sub-project whose goal is building a classification algorithm to estimate the surface type (paved vs unpaved) of each road of Greater Maputo from public satellite images. Indeed, among the 12.9 thousand km of roads in the area only 5% are known to be paved or unpaved. In detail, the solution is developed in the framework of Object-Oriented Data Analysis [1] with each road being mathematically modelled as a cloud of random points (i.e., image pixels) in the RGB space.