Fuzzy Logic (FL) is a qualitative approach based on approximation reasoning that is close to human thinking. A fuzzy system (FS) is a structure that represents inputs into the output universe of interest through fuzzy logic principles. In FSs, both subjective and objective inputs, which can be both numerical and linguistic data, are in consideration. It has been very popular for more than forty years in transport engineering applications such as speed control on expressways [19], signalization for traffic control [20,21], seaport [22] and transit [23] operations, lane-changing simulation models [24] and congestion-related applications [25,26,27,28]. In [25], the authors measured the level of congestion by using the same fuzzy approach with inputs such as speed reduction rate, the proportion of delay time within total travel time, and traffic volume to road capacity. Patel and Mukherjee [26] classified the traffic according to a fuzzified index of congestion and the average speed level on the urban road network. Here, the congestion index was calculated as an output of the relationship between actual and free-flow travel time. The authors showed that the fuzzy approach was better at showing the real congested situation than other traditional congestion index values. Another fuzzy congestion evaluation study considering average speed as the input variable was presented by Hamad and Kikuchi [27]. They used travel speed, free-flow speed, and the proportion of very low speed in the total travel time as input variables to determine the congestion situation.

Additionally, in [28], Kikuchi and Chakroborty studied a fuzzy approach for handling the uncertainty embedded in the definition of the level of service (LOS). They criticized the current HCM procedure, arguing that it does not accurately represent the notion of LOS as a user-perceived measure, and questioned whether a single measure (e.g., density) could capture all of the factors that affect LOS. Thus, they provided a framework that handles uncertainty under the different paradigms: deterministic, probabilistic, or possibilistic. Further studies in this field include [16,29,30,31]. On the basis of the described analyses, it is indicated that fuzzy-based applications have a preferable performance. However, recent experiments have generally been focused on detecting and testing abnormal events of traffic, and have rarely addressed the real-time estimation of the traffic state of the network.

The idea of detecting traffic congestion using the Mamdani-based fuzzy approach has already been studied and proven to be effective [32,33,34]. In [32], the authors used three inputs—the length of the lanes, the number of lanes, and flow data—to obtain the congestion level output. However, the experiment was only based on a one-week period of data; Kalinic and Keler [33] worked on a fuzzy method that compares two input sets: flow–density, and occupancy–mean speed parameters for detecting traffic congestion; Kalinic and Krisp [34] presented a model containing only two inputs (flow and density), as in most classical traffic studies, which relate pairs of fundamental variables, with a few noticeable exceptions, like the application of Catastrophe theory to a 3D traffic state space, introduced in the 1980s [34], and which is still being studied and applied for the identification of traffic congestion [35,36].

With respect to the contributions of those reference works, we believe that traffic congestion should be evaluated in terms of speed since it is a function of speed reduction in time. There are many studies considering speed as a parameter with or without others to determine the level of congestion on expressways and urban roads [8,25,26,27,29,31]. Specifically, in [8], the authors stated that a speed-based threshold has a greater impact on congestion than a threshold based on capacity does. From this point of view, we focused on the speed values. In this paper, we relate the classification of traffic states to the average speed variable instead of the density in order to obtain a perspective more compliant with the DATEX II European standard, which focuses on providing information to drivers, rather than being directed towards planning purposes. However, we also acknowledge that the extent of congestion is multi-dimensional, and the use of a single variable cannot give a decent assessment [37]. It is difficult to assume that there is a single value determining the entire traffic situation [38]. Furthermore, congestion is a state of traffic flow characterized by the fundamental variables together, which must be considered as a whole in the classification operation. Therefore, we developed a two-stage traffic congestion detector that is able to predict speed values by considering flow and density values, and which then provides a qualitative estimate of congestion that drivers will find more trustworthy, according to this reasoning.

With the aim of suggesting that a fuzzy approach might be more convenient than a deterministic one for catching the inherent uncertainty in the drivers’ perception of traffic congestion when the flow characteristics are highly perturbed in oversaturated conditions, we tested our approach using data collected from the network on a motorway in Italy over a period of more than 8 months. The two-stage detector was able to predict the short-term speed values according to fuzzy rules and then classify the corresponding traffic states based on the EU DATEX II standard ranges.