**3. The hypermode concept**

Currently, mode choice is considered a planned behavior and is embedded within traffic assignment procedures only in a static context [83], which obviously does not capture the influence of any real-time events. With regard to dynamic modelling, mode choice is usually considered a fully pre-trip behavior. This paper

*Models and Technologies for Smart, Sustainable and Safe Transportation Systems*

interpret the decision-making process [32, 68–72].

sectional data [57–59].

Finally, different attempts have also been made to model an habitual behaviour (holding decision), but taking into account temporal correlation for the same user, thereby showing how tastes can vary for the same traveller using short-term cross-

The pre-trip and habitual choices have been extensively investigated in the literature; however, some scholars have also dealt with the explicit simulation of mode choice dynamics with regard to both short- or long-term scenarios. In particular, [60–63] study short-term mode choice dynamics using discrete choice method and panel data. With regard to long-term mode choice dynamics, [64] investigate commuting behavior within the traditional maximum utility framework, whereas alternative approaches have tried to take into account more complex behavioural determinants and processes such as habits and learning. In particular, [65] derived decision rules based on neural networks to predict activity scheduling and mode choice; [66] developed a computational process model to mimic travel decisionmaking process; [67] developed an agent-based process to simulate travel behavior in terms of information acquisition, learning, adaptation and decision heuristics. Recently, the Markov chain approach has been fruitfully adopted to model and

With regard to the within-day travel mode choice behavior, it can be assumed that a typical traveller chooses a transport mode (or a combination of transport modes) and may change his/her initial choice by switching to other modes before leaving the origin and/or during the trip. Obviously, such behaviour is reasonable only in a multi-modal or inter-modal context. On the one hand not many contributions can be founded in the literature; on the other hand the landscape of available mode options is evolving particularly at the urban level. Multi-modal networks are rapidly growing, and a new generation of mobile, personalised information systems and intelligent transport systems are ready to support this flexible and adaptive behaviour by provid-

ing assistance in the planning and implementation of multimodal trips [73].

considered in the hypermode approach proposed in this paper.

As a matter of fact, an increasing number of users may reconsider their initial travel choices. However, not many contributions can be found in literature. From a psychological viewpoint, the study undertaken by [74] considers a two-level approach to simulate the mode choice. At the first level (more related to the person) the authors apply the comprehensive action determination model, which assumes that intentional processes, habitual processes, and normative processes lead to a certain level of propensity to use the private car. The second level of choice (more related to the trip) is characterized by situational influences, where trip purpose, disruptions on public transport, and weather are identified as predictors. The authors conclude that the multi-level approach is a promising alternative to conventional models. These insights from the field of psychology are valuable for the correct interpretation of the decision-making processes of travellers, and will be

Different contributions have analysed mode choice as a sort of path choice in a broader context of a multimodal network [49, 75–77], by considering interconnected networks, one for each different transport mode. Such an issue has been addressed in a multi-modal context through the well-established supernetworks [78, 79]. However, such an approach has been mainly adopted/used for modelling elastic demand assign-

Contribution by [80] take the mode availability into consideration, but mainly from an assignment perspective as both model the user decision to change mode at each node. In [81] each option (mode and route) is associated with a probability of immediate availability, which is one for private modes and less than one for public transport, the latter being a specific value affected by service frequency. The author therefore revised the transit assignment problem by taking into account

ment problems; it is not very flexible to address possible adaptive behaviour.

**8**

investigates an adaptive mode choice behaviour and presents the results of an empirical study undertaken to validate the approach. It focuses on the potential effects of real-time events on both pre-trip and en-route mode choices.

For reason that will become clear below, this adaptive mode choice will be hereafter called "*hypermode"*, in analogy to the hyperpath concept proposed for the route choice in transit assignment [84]. The hyperpath approach suggests that travellers first identify a set of attractive lines that connect their origin–destination (O-D) pair; then they choose a specific service according to certain strategies. Such strategies can be based on the minimization of travel time, waiting time, walking distance, or the number of transfers A more complex strategy can also consider the influence of real-time information on path choices [85, 86]. In an analogous way, the hypermode concept stipulates that travellers identify a set of feasible modes for their target trip and may make their final decisions later based on real-time events. These adaptive mode choices have been recently facilitated by the development of Information and Communication Technologies (ICT) such as smartphones, as well as Intelligent Transport Systems (ITS) such as vehicle tracking and prediction. For example, travellers can now make informed mode choices based on estimated time of arrival of buses/trains/trams, or the availability of shared bikes at any given docking station. Such adaptive travel behaviour is suitable for dense urban areas, where plenty of mode options and access points are available to travellers, and walking is always an option especially for short trips. Given that 50% of the trips in urban areas in Europe are shorter than 5 km [87], the hypermode concept enjoys wide empirical support. This extra modelling dimension could lead to a significant yet challenging advancement in the modelling of multimodal transport networks.

This section illustrates this notion by proposing a conceptual analytical framework along with a few examples.

#### **3.1 Decision-making architecture**

In this section, we formally introduce the *hypermode* concept, which is analogous to the hyperpath concept proposed for the route choice in public transit assignment [84]. The hyperpath approach suggests that a traveller first identifies a set of attractive lines that connect the origin–destination (O-D) pairs. Then, he/she chooses a specific service according to a certain strategy, which can be based on the minimization of travel/waiting time, amount of walking, or number of transfers. A more sophisticated strategy can also take into account the influence of real-time information on path choices [86]. In an analogous way, the hypermode concept stipulates that travellers identify a set of feasible modes for their target trip and may later make their final decisions based on real-time events. These adaptive mode choices have been recently facilitated by the development of Intelligent Transport Systems (ITS), and Information and Communication Technologies (ICT).

The underpinning decision making process involved in the hypermode concept is articulated in two levels.


**11**

*Adaptive Travel Mode Choice in the Era of Mobility as a Service (MaaS): Literature Review…*

shared modes), weather conditions (relevant to walking and biking), vehicle arrival time information (relevant to scheduled or unscheduled public trans-

This adaptive behaviour can occur at the following different stages of the trip:

• The user has not yet left the origin and has a set of modes in mind that could bring him/her to the destination with acceptable time and cost. Just before leaving the origin the user reassesses these modes based on real-time events such as weather, real-time bus information and so forth, which may influence

• The user has just left the origin with a specific mode in mind (e.g. tube). He/ she then approaches a tube station and notices a disruption or heavy crowding,

• The user has chosen an *access point*, which is a specific location where he/she can access several modes that can all serve the trip. The user approaches the access point and then chooses a specific mode based on a combination of his/ her preferences (e.g. first coming/least walking/least transfers) and real-

The extent to which the real-time events affect the mode choices varies among

**Table 1** illustrates the proposed approach and a non-exhaustive list of factors

The realization of a specific mode choice is therefore the consequence of the mode first belonging to the feasible set (choice level 1), and then actually chosen

**Figure 1** illustrates, in further detail, individual components of the decision

Any of the traditional mode choice models can be applied to calculate the probability at the first level. Once the probabilities of all possible modes are calculated, the set of feasible modes can be formed, which is a quite standard procedure and thus omitted here. In the second level of the decision making process, the feasible modes are subject to re-interpretation and their probabilities are reassessed based on real-time events. For example, if walking is the preferred mode with the highest probability at the first level, and the weather is rainy in real time, the probability

The whole procedure may be easily formalized in a compact formulation coherent with existing assignment models, thus may be implemented for simulation any

individuals. For example, some users may take their preferred modes in any circumstance. This is particular the case for travellers who use their own vehicles, such as private cars or bikes (cyclists who use their own bikes usually stick to the same mode in case of very adverse weather conditions). Such behaviour is referred to as *planned* in this paper since it is not adaptive, and only involves the first level of the decision making process. Such planned behaviour can be subsumed by the proposed two-level decision making paradigm, as it is a special case with decision parameters on the second level being rigid and non-responsive to

hence immediately considers another mode from the feasible set.

the user's the final choice of mode within the feasible set.

*DOI: http://dx.doi.org/10.5772/intechopen.98432*

time events.

real-time events.

**3.2 Factors affecting adaptive mode choices**

associated with walking decreases.

transportation system (see technical report [88]).

affecting choice probabilities on the two choice levels.

within such set with given real-time events (choice level 2).

making processes with inputs and outputs of the two levels of choices.

port), and disruption or crowdedness.

shared modes), weather conditions (relevant to walking and biking), vehicle arrival time information (relevant to scheduled or unscheduled public transport), and disruption or crowdedness.

This adaptive behaviour can occur at the following different stages of the trip:


The extent to which the real-time events affect the mode choices varies among individuals. For example, some users may take their preferred modes in any circumstance. This is particular the case for travellers who use their own vehicles, such as private cars or bikes (cyclists who use their own bikes usually stick to the same mode in case of very adverse weather conditions). Such behaviour is referred to as *planned* in this paper since it is not adaptive, and only involves the first level of the decision making process. Such planned behaviour can be subsumed by the proposed two-level decision making paradigm, as it is a special case with decision parameters on the second level being rigid and non-responsive to real-time events.
