Summary
Cities worldwide face the dual challenge of increasing density while developing sustainably. These pressures strain transportation systems, necessitating optimal performance. In the Netherlands, transport policy focuses on three main objectives: accessibility, liveability, and safety. This thesis addresses accessibility and liveability.
Liveability concerns the fit between residents and their living environment. This fit is an ecological relationship involving interactions with the built, natural, and social environments. The thesis explores two approaches to operationalising this ecological fit: static and dynamic. The static approach, exemplified by tools like the Leefbaarometer, views liveability as a measurable outcome. It considers the valuation of the environment as a proxy for fit, obtained through surveys or hedonic pricing methods. This approach is widely used in current transport policy evaluation but is limited in its ability to do ex-ante analysis.
On the other hand, the dynamic approach views liveability as an ongoing process. It considers fit the difference between what a resident expects to observe and experiences. This approach uses generative modelling, which involves causal probabilistic relationships. The manner in which these generative models are updated is unique, attempting to capture the dynamics of the reciprocal relationship between the resident and the living environment. The resident embodies the generative model, containing both internal and external states. The generative model is definitive of a resident's niche. Construction of the niche is afforded by accessibility.

Objective
The thesis aims to work towards future operationalisation of the dynamic approach to liveability. While full implementation is far beyond the scope, the objective is to understand the role of representations in defining and operationalising liveability. In the context of transport systems, representations can be understood as simplified and compressed digital twins of complicated urban environments. These representations can be used for decision-making in infrastructure planning and policy formulation.
The dynamic approach to liveability relies on the action-perception loop as its generative model. This loop means that perception is anticipated, an action in itself. Selecting optimal actions is a matter of maximising epistemic and pragmatic value. Writing a thesis is an action; hence, we follow the formal decomposition of action selection:
• Epistemic component: develop a theoretical framework, exploring the action-perception loop in liveability and the role of representations within this loop. Bridging cognitive science and transport modelling enables novel machine learning applications.

• Pragmatic component: automate the operationalisation of liveability by applying urban representation learning techniques to the province of South Holland. Our novel contribution is the developed mathematical model based on spatial convolutions. The study uses various data sources relevant to transport planning, including road networks, public transport schedules, aerial and street view images, and points of interest.

Modelling Study
Addressing the pragmatic component, we develop a novel urban representation learning technique using the H3 hierarchical hexagonal discrete global grid system (DGGS). Neural networks are used to extract representations such that each discrete hexagon is assigned a coordinate in the metric representation space. Features from different data sources are extracted using individual encoder networks. These features are then combined using a late-fusion network to obtain the final representations. We develop Ring Aggregation, a mathematical model to fuse multiple features while accounting for spatial context through sampling heuristics and spatial convolutions. The methodology considers insights from the dynamic approach to liveability, like landscapes of affordances, analogous to location-based accessibility. No operationalisation of the dynamic approach is developed in this study. Instead, this study applies the static approach and evaluates urban representations using multiple univariate linear regressions with Leefbaarometer scores as targets.

Key Epistemic Observations
• The action-perception loop is central to the dynamic approach to liveability but absent in the static approach.
• Representations play an instrumental role in the action-perception loop, informing action selection.

Key Pragmatic Observations
• Regarding sampling heuristics used to train the late-fusion network, Euclidean distance and location-based accessibility perform similarly, better preserving urban area integrity.
• The configuration of Ring Aggregation significantly impacts urban representation quality, showing heterogeneity between Leefbaarometer scores. Socially oriented scores can perform just as well with smaller receptive fields and steeper weighted average functions like exponential.
• Different data sources excel in predicting various aspects of liveability, suggesting the importance of integrated data approaches in transport policy.
• Compared to urban2vec and M3G studies, Ring Aggregation outperforms them in predicting Leefbaarometer scores across the board.

Looking Forward
By reframing the roles of indicators, perception and preferences, this thesis concludes that static and dynamic approaches to liveability are complementary. Complementarity implies that the static approach to liveability can be used to bootstrap the development of the dynamic approach. While 'living' digital twins operationalising the dynamic approach do not yet exist, this study provides building blocks towards their development. Future directions for research and application in transport policy centre on the hypothesis that hierarchical active inference models are already transport models, just waiting to be used.

Human understanding and interpretability of AI models have become a growing problem in policymaking, particularly in urban planning. As AI models become more complex and opaque, decision-makers struggle to translate and explain their outcomes into human terms, leading to policy models that lack explainability and human-understandable outcomes. This research explores whether applying Explainable Artificial Intelligence (XAI) techniques, specifically Local Interpretable Model-agnostic Explanations (LIME), can enhance the interpretability of computer vision-enriched discrete choice models (CVDCMs) for street view images in urban analysis. CVDCMs analyze the liveability of areas using street view images, by analyzing the full image to determine its liveability score. This is unlike traditional urban models using object detection for street view images (relating liveability to a certain set of predefined objects in street views and measuring feature importances), thereby introducing a bias of the modeller in the object set choice. CVDCMs, in combination with XAI, can avoid this specific modeller bias, providing a more holistic view of human perception of urban liveability and improving the CVDCM’s utility for policy advice by possibly providing new insights into urban liveability.

The methods employed in this research include applying CVDCMs to perform a face validity analysis with street view images to gain an initial understanding of the model’s decision behaviour. LIME was then used to generate explanations for these model decisions by segmenting the images, perturbing these segments to create modified images, and analyzing the impact on the classification of the street views into ’good’ or ’bad’. This approach highlights which ’parts’ of street view images contribute most to preferred liveability, thus providing a more comprehensive view of human perception. LIME can showcase the complex decision behaviour of the CVDCM by considering the entire street view image as a possible explanation for a liveability score, without requiring a predefined set of factors. This contrasts with non-XAI methods, which leave users without insight into how the model arrives at its
results. By not relying on prior object detection, LIME avoids modeller bias and allows for a more holistic view of human perception, enhancing the interpretability and transparency of CVDCMs. The LIME segmentations were evaluated with ground truth images and the LIME explanations were assessed with various metrics (Binary Classification Ratio, Coefficient of Variation, and Probability Distribution Uniformity Metric), ensuring robust and reliable interpretability of both the segmentations and the LIME explanations.


The study revealed several significant findings. The quality of LIME explanations is highly dependent on the segmentation process, as already also found in earlier research. Street view images, which contain a variety of complex objects, present challenges in achieving meaningful segmentation. While ground truth analysis can improve segmentation quality, it is labour-intensive and prone to inconsistencies. The lack of semantically meaningful segmentation results in non-human-understandable explanations of the model’s outcomes, and thus does not increase the interpretability of CVDCMs. Additionally, the sampling process of LIME results in perturbed images that significantly differ from the original image.
This impacts the utility-based liveability score and necessitates a unique classification threshold for each image in the deterministic classification, complicating comparisons and reducing explanation coherence. In the probabilistic classification, this is not necessary, but the distribution of the classification probabilities is therefore not adequate, resulting in scientifically less valid LIME explanations. Moreover, the computational intensity of the LIME methodology limits its scalability for analyzing large datasets necessary for comprehensive urban policy analysis. Therefore it is concluded that in the current applied methodology, the interpretability of the complex model is not increased and no human-understandable explanations are created. This research uniquely identifies the complications of using CVDCM with LIME image analysis and suggests that simpler object detection models combined with LIME tabular explanations could offer a valuable alternative.

In response to these findings, several recommendations are proposed to increase
interpretability and create human-understandable explanations. The use of street view-trained segmentation models for superpixel creation (via object detection) can provide human-understandable LIME explanations by providing semantically meaningful segmentations, which could also decrease the computational time of the LIME method, thereby introducing some bias but still enabling the analysis of the holistic view of human perception. Further research on the distribution of utility scores in the sampling process and experimenting with different parameters or distributions for the sampling process could improve the LIME
explanations. Additionally, applying methods to integrate regression models directly within the LIME framework, rather than converting them to classification models, could avoid the pitfalls of thresholdbased classification. These recommendations ensure that the explanations are more aligned with the process of urban policy advice creation, as they become logical and actionable through these improvements, enabling its use for municipal policy advice.

This paper investigates cycling route preferences, with a focus on the cycling environment. To represent the cycling environment street-level images were used. A recently proposed model incorporates computer vision into a traditional discrete choice model to accommodate choice tasks involving numerical attributes and images. This computer vision-enriched discrete choice model (cv-dcm) was applied using a stated choice experiment, where respondents had to choose between two cycling routes. Each route was defined by three attributes, including commute time, number of traffic lights, and the cycling environment, the latter visualised using street-level images. While the cv-dcm relies on a neural network, making interpretability challenging, this study addressed this by collecting detailed cycling environment attributes. Results showed that the cycling environment was the most influential factor, with cyclists preferring green areas and separated cycling lanes. On average, cyclists were willing to take a 1.5-minute detour for a cycle trip of 11 minutes to use a separated cycling lane instead of a mixed-traffic road. These insights offer valuable insights for policymakers aiming to design cycling environments that align with cyclists’ preferences.

Transportation systems are pivotal in shaping the economic and social dynamics of contemporary societies, fostering connectivity and opportunities while reducing geographical distances. Despite these benefits, they also contribute to adverse effects such as emissions, congestion, and traffic fatalities. Effectively developing and maintaining transportation infrastructure and services that cater to evolving population needs and align with environmental goals requires accurate forecasting of travel demand. However, due to inherent uncertainty in individuals' behavior and data limitations, forecasting this demand is a complex task.
A common limitation often encountered in transport datasets is class imbalance, with regard to the utilization of the different modes. Class imbalance in this context refers to the uneven distribution of samples among the various modes. Modes with a higher number of samples are termed majority modes, while those with fewer instances are labeled as minority modes. The existence of class imbalance within the dataset has the potential to compromise the performance of classifiers, especially for the minority modes, leading to inaccurate forecasts. This, in turn, may result in insufficient investments and provisions for these modes, ultimately having adverse consequences for the population segments that rely on them. Existing studies in the literature have either entirely overlooked or only partially addressed the impact of class imbalance. Recognizing the significance of precise demand predictions and acknowledging the identified gaps within the literature, the primary research question of this study revolves around systematically identifying and addressing the impact of class imbalance in mode choice forecasting.
To address the main question, a framework was proposed. This framework encompassed various aspects including a) the measurement of class imbalance within a dataset and the assessment of its impact on classification performance, b) the investigation of other challenging factors coexisting in imbalanced datasets, with a specific focus on class overlap, and c) the proper evaluation of classification performance across classes. As an integral part of this framework, the 'Performance Gap Metric’ was introduced - a metric employed to evaluate the difference in classification performance between the majority and minority classes. Establishing a threshold of 20%, favorable classifier performance was determined when this metric fell below the threshold, signifying the classifier’s equitable treatment of both minority and majority classes. Subsequently, this framework was applied using the ODiN data as a case study to predict mode choices in the Netherlands. Mode choices encompassed car, bike, and transit, with car representing the majority and transit the minority class. Two modeling techniques, namely Random Forest and an MNL model, were employed in conjunction with various sampling techniques, including the SMOTENC, the Neighborhood-based Undersampling, and the Separation scheme...

Flying has become increasingly accessible in the last few decades, leading to enormous growth in aircraft traffic worldwide. Amsterdam’s Schiphol Airport is the fourth-busiest airport in Europe based on passenger traffic, and noise pollution from the airport affects over 63,000 individuals daily. This study aims to determine if the presence of trees has the potential to significantly mitigate noise pollution from aircraft across seasonal leaf patterns and across different configurations in a simulated street canyon near a major airport. Thirty-six adolescent Common Linden trees were placed in a simulated street canyon near Schiphol Airport’s Kaagbaan runway, where sound, weather, and flight data were collected between February through May. Two additional configurations of the trees were also tested to evaluate the effect of planting density and patterns on scattering and reflecting noise. Trends in sound pressure levels measured inside the street canyon were compared to levels measured by a reference microphone, and a linear regression analysis was performed to determine the effect of weather and trajectory variables on the differences in sound pressure levels between these two environments. Between 0.68 and 3.3 dB of noise attenuation were observed in the experimental courtyard for arriv- ing flights, versus between -2.65 and 0.5 dB of noise attenuation for departing flights. Furthermore, while around 10 percent (R2 =0.099) of variation in the noise attenuation
of arriving flights could be explained by flight trajectory and weather variables alone, this percentage was significantly higher for departing flights (R2 =0.46). These results are in line with previous research which found that the interaction of building properties with meteorological variables and flight trajectory have the most influence on sound propaga- tion of aircraft noise within a street canyon environment, but also suggest that vegetation can play a role in mitigating noise pollution. Further research is required to determine if the presence of adult leaves or the psychological effects of greenery on the human per- ception of aircraft noise pollution could augment the modest noise pollution attenuation effects of trees seen in this experiment.

European cities are implementing diverse strategies to curtail car usage. Understanding the impact of these policies necessitates insights into mode choice behaviour. However, for conventional discrete choice models, utility specifications must be defined upfront, potentially leading to misleading policy recommendations. This problem is solved by Supervised machine learning (ML) models. However, they are challenging to interpret, which is crucial for evaluating transportation policies. We employ Shapley additive explanations (SHAP), a model- agnostic explainable artificial intelligence (XAI) tool to address this gap. The main advantages of SHAP are its foundation in game theory, the ability to highlight individual taste heterogeneity and non-linear effects. This paper aims to shed light on the potential of SHAP to improve current transportation mode choice models. Using a random forest (RF) model with the TreeSHAP estimation method, we compare SHAP insights with those derived from a traditional multinomial logit (MNL) model. The results indicate that SHAP can detect the absolute importance of features. Substantial preference heterogeneity for car choice is perceived for features reducing car usage, as opposed to features increasing car usage. Non-linear effects, such as reciprocal functions and clustered patterns, are observed for certain features. MNL and RF models disagree on the importance and heterogeneity of features, and the MNL model fails to model highly nonlinear effects. For policymakers, insights suggest that increasing parking fees and promoting car sharing may be feasible options. However, the efficiency of these measures may vary due to preference hetero- geneity. The results underscore the need for further investigation into the reasons behind the different model results and different behaviour notions of SHAP, MNL and RF.

About 75 percent of CO2 emissions caused by the combustion of fossil fuels stem from cities. Climate policy minimising the amount of emitted carbon can not be evaluated without accurate quantification of CO2 emissions.
Not only the amount of emitted carbon is of interest, but also being able to place emissions in both the spatial as well as the temporal context is necessary. This gives policy makers and scientist the ability to compare the climate impact of neighbourhoods and cities. For doing so, it is necessary to have a rasterised emission estimate with a high resolution.

This study proposes a new method using a Random Forest model trained on the Hestia dataset. The model utilises spatial co-variates, such as population density, night time lights, to estimate emissions in a resolution of 1 X 1 km.
To assess the spatial generalisability of the model, the emission observations of the Hestia dataset are compared against model estimates. The explained variance and R2 is the highest in log transformed form at ∼0.47. The features and hyper-parameters
selected through this method are found to be stable under a variety of model specifications. This provides evidence that the presented model performance generalisability are the that could be achieved given the current data. The relevant covariates for the model are the following, in order of importance: Nightly radiance, intersection density, population, GDP, PM25 and Pedestrian intersections. Inspection of the partial dependence shows that each of these variables, except PM25, have a positive increasing relationship with the emission estimate. This relationship quickly converges for higher values. The highest relative errors mainly occur around the peripheries of cities, for cells with low emission values.

This study lays forward a new scalable approach to estimate emissions and the spatial distribution in urban areas. Currently the study data is limited, but as more data becomes available a Random Forest regression approach could prove to produce accurate insights. Also the findings in model design and evaluation could be used in future studies. For city planners the results can be used to gain a first insight in the spatial distribution of emissions within their city. Pinpointing high emission zones or comparing different urban areas. Knowledge of the spatial distribution of cities can be a pathway of understanding how to develop less carbon intensive cities.

More and more research shows the substantial health repercussions
of air pollution. Therefore, improving air quality is high on political agendas in modern societies. In the Netherlands, particularly around major roads, NO2 standards set by the government are often exceeded. Air quality models are used to monitor air quality values and design policies to reduce air pollution. Currently, authorities in the Netherlands use a Gaussian Plume Model for decision-making, but this model paints a rather skewed view of reality due to
its underlying assumptions. This research contributes to academic
knowledge about air quality modelling by evaluating two innovative model types, a physics-based LES model and a data-driven regression model, for their usage in decision-making to improve air quality. This is done by comparing the performance of both models with the performance of a Gaussian Plume Model for predicting NO2 levels around a large highway in the Netherlands. Also, two
combinations of the LES model and the regression model are examined. It is concluded that both the LES model and the regression model show potential for accurately predicting air quality around highways in the Netherlands. The LES model is particularly suitable for predicting high NO2 levels, and the regression model is considered suitable for predicting the average NO2 levels over a longer timeframe. A model in which the LES results were combined with a
regression model outperformed the original models and is therefore considered to hold the most potential for usage within air quality
policy.

In the recent years, police violence and discrimination towards minorities have become widely criticized by news outlets and within society. The trust in the police has decreased over the last years as ethnic profiling and the deaths of multiple people of colour have come to light. The police law enforcement and their officers have been criticized for their behaviour, as there has been a disproportionate number of arrests targeting people of minority. However, studies researching the occurrence of discrimination in police law enforcement and its effects, have not been conducted in The Netherlands. This study aims to identify which factors influence a police officer’s decision, as this arguably has the most impact on a citizen, and whether discrimination can occur in these factors. The first step is to identify the factors, which is achieved through a literature review. As not much research is performed in The Netherlands regarding discrimination in police law enforcement, multiple interviews are held with police officers to verify the obtained factors from the studies. This resulted in five factors being identified, namely type of crime, ethnicity, gender, familiarity with the police and group size. The above mentioned factors are used in the survey, which was distributed among the police officers stationed in the bureau Haarlem, North-Holland. The factor ethnicity influences a police officer’s decision for citation. However, contrary to the results obtained from the literature, a Non-Western individual is not more likely to be arrested compared to a Western individual. This study found the exact opposite; an individual of Non-Western ethnicity is less likely to be cited compared to a Western individual. The relative importance of this factor is however very small as it contributes for less than 5% on the decision of a police officer for citation. The other factors that influence a police officer’s decision for citation in this study are mixed gender and type of crime, with a relative importance of less than 8% and 87% respectively. The factors group size and familiarity with the police are not statistically significant, which means that they do not influence the decision of a police officer. There are multiple limitations in this study. First, this study was conducted with a limited number of respondents and hence it is hard to generalize to the population. Secondly, the study was conducted with police officers stationed only in Haarlem, therefore hard to generalize to the police officers in The Netherlands. Thirdly, the results may have been exposed to hypothetical bias. Lastly, the research method can be seen as a limitation as this method is used for consumer preferences and not to explore moral considerations. Lastly, this study provides four recommendations for further research. Firstly, more research must be conducted with regard to discrimination in police law enforcement. Secondly, respondents participating in the study should be increased and multiple bureaus should participate to be able to generalize the results. Thirdly, future research should incorporate the police officer’s educational level. Individuals, who have experience in critical thinking, rely less on profiles in their decisions. Lastly, by identifying heuristics and shortcuts, and researching whether these methods contain possibilities for discrimination to occur, is valuable knowledge to obtain to succeed at the aim of policing for everyone.  

The European Commission recently published a proposal for an Artificial Intelligence (AI) act that requires the development of trustworthy AI systems for European Union markets. The proposal clearly mentions that AI systems should make use of Explainable AI tools to increase transparency and interpretability.Financial institutions in the Netherlands are required by law to detect and monitor fraud in their infrastructure. Fraud detection in Financial Services (FS) or the FinTech industry is increasingly performed by Machine Learning (ML) and Artificial Neural Network models that provide high classificationperformance. ML/ANN-based fraud detection systems that are necessary for maintaining trust in the Dutch financial system are classified as high-risk applications by the proposal for the EU AI act. The EU AI act will directly impact high-risk AI applications used within the EU markets, Therefore,the Dutch financial institution sponsoring this research wants to future-proof their ML-based fraud detection to improve transparency and trust by solving the model interpretability problem. Explainable Artificial Intelligence (XAI) is a domain of AI research that seeks to solve model interpretabilityproblems of black-box ML models. In this thesis research, proofs of concepts are demonstrated for the investigation of two XAI approaches - TreeSHAP & Diverse Counterfactuals to improve the model explainabilityor interpretability of ML/ANN-based fraud detection systems. This research pioneers the investigation of Diverse Counterfactuals to improve model interpretability in ML/ANN-based frauddetection systems. Based on the existing literature, this is the first instance of research investigating Diverse Counterfactuals for generating explanations to ML/ANN-based fraud detection models trained using synthetic transaction datasets. Before demonstrating the proofs-of-concept, an extensive literature survey has been conducted to map the XAI research landscape, formulate an XAI taxonomy, and conduct a comparative analysis of XAI approaches to select and describe in detail, the relevant approaches for the use-case at hand. Subsequently, several ML and ANN models have been trained and tested using the PaySim synthetic transaction datasets. To overcome model performance challenges due to data quality issues and high class imbalancein the datasets, several experimentation scenarios involving hyperparameter optimization, SMOTE Oversampling and class-weighting have been investigated. Subsequently, two high-performing models (XGBoost & MLP) from these experiments have been used to demonstrate the proofs of conceptsby investigating TreeSHAP and Diverse Counterfactual algorithms. TreeSHAP algorithm greatly improved the interpretability of the global and local model behavior of the XGBoost-based fraud detection models. Diverse Counterfactuals algorithm-generated diverse but unfeasible counterfactuals. Counterfactual algorithms suffer from computational inefficiency and therefore, further research has to be conducted to generate feasible counterfactuals. Future work on model explainability should also conduct a human-grounded evaluation of the explanations to evaluate the quality or goodness of the explanations. Finally, real-world transaction datasets should be used instead of synthetic datasets so that the research is generalizable to the real world

On July 22nd 2018 the Noord/Zuidlijn, a metro line crossing the city center of Amsterdam, became operative. This entailed a transformation from a direct bus and tram network to a network consisting of a metro trunk line and bus and tram feeder lines. This study explores and uses interoperable smart card data from before and after the Noord/Zuidlijn became operative to measure the regional travel time and transfer impacts of this network overhaul. On a working day on average 1,350 hours of travel time are saved. The travel time savings and losses are 2,350 and 1,000 hours per day respectively. Additionally, 2,500 extra transfers are made per working day. The transfers gains are 4,500 transfers and the transfer losses are 7,000 transfers per working day. 20% of the travelers experiences a decrease in travel time of more than 1 minute, 10% experiences an increase in travel time of more than 1 minute. Furthermore, interoperable smart card data showed to be promising as it captures travel behavior across multiple operators. However, the fact that the number of travelers is given in bins complicates analysis. This study develops a methodology to work with interoperable smart card data. Additionally, as only few studies have evaluated the transportation impacts of a large-scale public transport network overhaul ex-post, the findings of this research could improve public transport planning and assessment.

Automated vehicles (AV) are expected to change our future way of transportation and bring many positive effects, such as better road/vehicle safety, reduced environmental costs but also increased on-board productivity. The commonly adopted approach to capture the effect of increased on-board productivity is to reduce the ‘penalty’ associated with travel time. This implies that people will show less aversion to longer travel times (i.e. congestion), prioritise arriving close to the preferred arrival time and thus, increase the peak congestion. However, a distinction should be made between the type of performed activity, since this could likely affect the departure time preferences and thus can have an impact on the congestion pattern. Previous work has shown that travellers with fully automated vehicles in which home activities could better be performed, preferred to depart earlier, and conversely travellers with automated vehicles in which work activities could better be performed, preferred depart later. In addition, the results suggested that automated vehicles could likely increase severe congestion in the future. However, these insights have been obtained with the use of a theoretical single link setting. Up to this point differentiation between activities has not yet been investigated for real-life road networks in which route choice can be incorporated. Therefore this research aims to investigate network effects of activity based departure time choice with fully automated vehicles. It uses the extended α-β-γ scheduling preferences by B. Pudāne and implemented these within a traffic simulation model. This model was assigned to a case study network of the ‘Haaglanden’ region. Results showed that travellers which use AVs to engage in home activities, increase congestion more to the beginning of the morning peak. Vice versa, travellers which use AVs to engage in work activities move congestion more to the end of the morning peak. Travellers which use AVs for both home and work activities would increase congestion in the middle. Considering route choice, a shift in vehicle-kilomtres (VKM) could be observed from the main road network to the underlying road network with the introduction of AVs. This has a negative impact regarding traffic safety, noise and air pollution. Although one might argue that future AVs will have increased safety levels and a more eco-friendly way of driving. Lastly, it was observed that with mixed traffic situations, non-AV users were ‘forced’ to shift their departure times out of the peak moments, move to less congested periods. Conversely, AV users were less affected by these longer travel times and prioritised their preferred arrival times. It is recommended that further research is conducted to include the dynamic modelling element of blocking back. In addition, the extended α-β-γ model parameter values should be assessed to arrive at more substantiated values. Moreover, it suggested to further investigate the impact of differentiation towards road capacities for AVs. Furthermore, this study did not investigate heterogeneity among travellers’ preferences, nor did it take into account endogeneity through which people may assign themselves to a certain type of activity. It is suggest to take these aspects into account in further research. This study assumed full automation, whereas it would be interesting to investigate the effects with lower automation as well. Lastly, an important extension of this research would be to also capture the effect of value of travel time differentiation with respect to route choice, as this study did not differentiate in probabilities to choose a specific route per vehicle type.

Hely has developed Mobility as a Service (MaaS) through multimodal Hubs,
which are offered within closed-user residential areas and made accessible
through an application. This exploratory study aims to assess people’s mobility
behaviour by the usage of Hely in a closed user group. It focused on two
aspects: the indication of (1) the community and (2) the use on the hubs by this
community. The community is characterized by mainly innovative bike- and
public transport-minded young professionals that seek a flexible and
convenient service. The motivation to use Hubs for its multimodality is
currently not widely supported by the community. Moreover, the Hubs reveal
travel behaviour with the shared (e-)car in its centre, independent of all trip
characteristics. Therefore, it can be concluded that currently a multimodal
MaaS-Hub mainly eases accessibility of cars for individuals that do not use
one often, which emphasizes the car as a lasting dominant factor within the
current mobility market.

Adaptive website content is an increasingly popular method for e-commerce platforms to personalize the content shown to its visitors in an attempt to increase the conversion rates of their platforms. However, in order to select the content relevant for a specific customer, the online platform should be able to rapidly interpret the explicit choices, as well as the implicitly sensed context, in which the choices are made. Creating a so-called context-aware e-commerce platform thus requires not only the understanding of users’ content-preferences, but also of the users’ context influencing this decision-making and the opportunities of the e-commerce platform to actively respond to this, in addition to analysis of customer behaviour that goes beyond traditional A/B testing used by websites and online services to quickly evaluate hypothesis with real users. Statistical choice modelling and machine learning are both methods with the ability of doing such advanced analysis of choice behaviour. By inferring preferences and trade-offs from people’s choices observed in real life, or hypothetically stated choice experiments, choice models can be estimated with which future choices can be predicted. As statistical choice models provide insight into the most important parameters that influence decisionmaking, conversion rates of e-commerce platforms can be maximized by optimizing those parameters. Moreover, it has long been acknowledged in discrete choice literature that the context in which a decision is made affects one’s decision-making. Machine learning on the other hand, takes an algorithmic approach and bases its prediction on patterns found in the data. The characteristic of machine learning to allow for high-order interactions that are not pre-specified, could be beneficial as the context in which decisions are made, is often fuzzy and highdimensional. However, as this is a black box method, the method will not explicitly give insight into the parameters influencing behaviour, but instead, train itself to maximize the outcome. Although a substantial amount of research is conducted on the difference between statistical choice models and machine learning, and A/B testing and machine learning are both widely used methods for the dynamic content selection on e-commerce platforms, and thus for conversion optimization, less research however, has been done on the complementary use of the two methods within the boundaries of a simple context-aware system-experiment. Nowadays, it is often stated that machine learning has an accuracy far beyond statistical methods and that it is the recommended method for future data analytics and prediction, as machine learning enables determination of outcomes in which large number of variables with complex relationships are involved. However, should machine learning always be the recommended method for an e-commerce platform trying to maximize its conversion rate with the use of dynamic content, or is more traditional statistical choice modelling sometimes still a better solution for smaller organizations less experienced in data analytics? The scientific objective of this study was to investigate the applicability of both statistical discrete response modelling and machine learning methods to maximize, separately or complementary, conversion rate on ecommerce platforms. By using the e-commerce platform Fatboy.com as test environment, in which both methods were required to handle the context-aware opportunities of contemporary online platforms, the potential in applicability of both methods could be examined. The objective within the case study Fatboy® was to understand how the conversion rates of e-commerce platforms can be maximized using statistical discrete response modelling, machine learning, or complementary to each other.

The population of The Netherlands is ageing. It is important to offer suitable transportation options for the elderly to keep them participating in social activities to prevent lonelines. This research investigated the preferences and usage of bus services of elderly in urban areas in The Netherlands and the corresponding heterogeneity. 256 elderly respondents were questioned by means of a stated preference survey. By using a multinomial logit model the values elderly place on the bus attributes were quantified. A latent class choice model was estimated to explore the heterogeneity and the nature of the heterogeneity in preferences among the elderly. Besides that, public transport card data from Groningen is used to compare theory with findings from practice. The research results show that elderly in the Netherlands can be divided into several classes with their own preferences. Not all elderly prefer a low distance to the nearest bus stop and most of the elderly prefer high-frequent bus options.

Deregulation within the aviation market led to constant decreasing revenue of airfares while airlines across the market spectrum sharpen their focus on ancillaries. The objective of this thesis project is to utilize revealed preference data in a case-study of KLM Royal Dutch Airways to better understand the customer relation towards ancillaries. Therefore, the willingness to pay for bundled flight ancillaries, also called branded fares, is estimated. Discrete choice modelling offers the opportunity to estimate the willingness to pay of the customers from the given data. A multinomial logit model was chosen and designed to enable the estimation of the parameter. Therefore, revealed preference data has been extracted, combined and cleaned with Python. A total of 47.350 flights have been analysed from 10 medium-haul destinations within Europe. For a better model fit the travel motive of the customers has been used to split the dataset into multiple actor specific constants being used to calculate the willingness to pay from. The results indicate far less willingness to pay for the ancillaries as currently asked for, namely 8.06 EUR for the Standard fare for business customers and 2.95 EUR for leisure customers. The Flex fare is estimated for 26.33 EUR for business customers and 4.82 EUR for leisure customers. The estimation result leads to the conclusion that it might be useful to implement price discrimination for the ancillaries as it is already in place for the airfare. Beyond the application on the willingness to pay, the revealed preference data has been analysed to reflect on the propensity and the amount of customer wish to pay for C02 offsetting. It shows that compared to stated preference data of other research the customers in this case-study pay far less frequently (only 1.02 percent of the passengers) and offset a tremendously less amount (on avg. 0.63 EUR) compared to the research based on stated preference.

This thesis describes a methodology for adding information about the preferences for new modes from external discrete choice models to exisiting transport demand gravity models. The methodology consists of 4 steps: (i) translation of deterrence functions towards discrete choice utility functions; (ii) linearization of non-linear translated utility functions, (iii) determination of factors to correct the scale of external models and (iv) the specification of the combined model. The thesis provides a case study related to "Urban Mobility as a Service" for the Eindhoven-Veldhoven area to show the practical validity of this methodology.

This Master Thesis document presents a research on trade-off analysis between security and efficiency factors during airport operations. Over the past few decades there has been an increase in security in the air transportation industry, with even more measures taken since 9/11. Besides affecting the operational costs of airports, the stricter controls also have been having an impact on the efficiency of the operations. Although that security and efficiency factors influence each other and on a frequent basis are being compared and considered in combination to take airport security operational decisions, the trade-offs made in order to take such decisions are not fully explicit. In order to make these security and efficiency trade-offs (which these decisions are based on) more explicit; a trade-off analysis formalized by Discrete Choice Modelling was conducted. Based on this analysis; implications were provided to evaluate, by the means of Agent-based Modelling, the effects of security and efficiency trade-offs made by airport security screeners on the operations of airports.