Investigating the Role of Points of Interest in Estimating Mobility Patterns in Cities: An extended Gravity model - London Rail

Urban cities are growing every day due to rising population, vehicular traffic, immigrants looking for opportunities and to accommodate this, cities are expanding and reshaping at an immense rate. People travel within a city for various purposes but the destination locations are the main reason for the movement. It is proved that locations influence a person's travel behavior. This research aims to study the influence of points of interest on human mobility available within a city and estimation of the travel flow between specified origin and destination locations. The city of London is chosen as the case study and the travel network is the London Rail. The research question this study aims to answer is "How to estimate human mobility by using points of interest?". There are sub-questions developed to guide the process of this study and they are used effectively throughout. In literature, there is not much research regarding intracity mobility and not much regarding estimating mobility with amenities/ points of interest (POI). There is research regarding mobility estimations at scales larger than cities and usually, it is travel flows from multiple origins to a singular destination. This research includes travel flows between both origin and destination locations. There are however implications in research that POI data can be used to quantify the influence of attractions on mobility and might prove to be better than the traditional gravity models. This research can help estimate the travel flows and aims to quantify the influence of points of interest in London. A new data preparation framework is designed from various data sources that lay the foundation for the data availability problem. The final dataset obtained contains the following variables: origin population, destination population, origin-destination distance, POI categories such as Commercial, Community, Educational, Entertainment, Financial, Government, Healthcare, Sustenance, and Transportation. For the model techniques, Negative Binomial and Poisson regression multivariate models are chosen. A series of experiments were conducted and the final experiments and the models within them were determined with the help of methods such as Akaike information criterion (AIC) and p-value. The model selection process was executed through a series of experiments to determine the most effective models. The model selection methods include the Akaike information criterion (AIC), p-value, Root-mean-square error (RMSE), and Coefficient of determination (popularly known as R-squared or R2). The RMSE and R2 methods were used to determine the best models among the sets of experiments created. Finally, model validation was conducted using the 'Sorensex Similarity Index' (SSI) method to quantify the similarity between the estimated and empirical data. The models were built separately for each of the days (MTT, FRI, SAT, and SUN) as time could not be included as a variable to the model. First, the process was executed for the day MTT and then it was replicated for the rest of the days. There were not many differences observed between the days as the POIs selected did not contain any difference between the days. If the timings of the POIs were available, then there could be clear differences observed between the days. Limitations and assumptions such as these were elaborated discussed throughout the thesis. The selected models are the first models for both Poisson and Negative binomial regression generalized linear models (GLM) and they perform better than the traditional gravity models. Though the differences observed are minimal, it is important to note that the Negative binomial models performed better on SSI while the Poisson models performed with the model selection methods such as RMSE and R2. The first models performed better using both the model selection and model validation methods. All the first models contain all the variables i.e. all the POI categories including the origin and destination populations along with the distance between them and the Differentiator variable. However, the SSI values are around 0.46 leaving a difference of 0.54 to reach the optimal estimations. This issue is discussed in detail and multiple methodologies including adding additional categorical variables such as economic/demographic indicators are suggested to overcome it. This research achieves the objectives presented in the paper and answers the main research question. The idea that POIs can be used to estimate mobility and they can be quantified within a city is novel and developing a new model satisfies and extends the research for both academic progression and policy implications. Regarding future work, there are limitless possibilities. The research can be extended for other cities, transport modes and also can be extended to other mobility models. The data availability problem is evident in the research and if that issue is not prevalent then the study could have been seamless. The POI categories are created using a subjective ideology and that can be improved using modern classification techniques. For policymakers and urban planners, this can help plan the city effectively and efficiently towards a sustainable future solving many problems such as ineffective traffic models, last-mile hike problem, congestion reduction, inequality minimization, irregular spatial designs, livability index, and much more. This study provides the foundation for a new thought process in Urban Science and it also discusses the possibilities this research could provide if extended further. Urban planning policymakers can understand the importance of amenities on human movement through this research and hopefully, allow them to make new creative decisions regarding urban planning. Keywords: human mobility, urban planning, intracity, amenities/points of interest (POI), gravity modelEngineering and Policy Analysi