The paper presents a methodological framework that integrates different data-driven techniques in order to detect the different traffic flow regimes (free flow, congested conditions, and so on) observed in signalized arterials as well as the manner traffic flow shifts from one regime to another (transitions). Traffic flow is determined by the joint consideration of the temporal evolution of volume and occupancy. The boundary conditions of the different regimes are identified via a fuzzy wavelet approach based on the volume-occupancy relationship. Moreover, a Bayesian network is developed in order to discover hidden associations between the observed shifts and the traffic flow conditions they occur. Results from the data-driven approach indicate the existence of four distinct traffic flow regimes; these regimes hold in arterials with different geometric and signalization characteristics. Finally, results are further discussed via an analytical model based on the kinematic wave theory; the comparative study of both approaches provides strong evidence that the presented statistical framework is in agreement with a simple and elegant analysis including traffic parameters that are observable and measurable.