The study of urban dynamics is to understand and analyse the dynamic properties of a city in the spatio-temporal domain. It includes the daily routines of the inhabitants, their movement patterns, geography, demography, and the social and economic condition of a city. Such data has created an excellent opportunity to integrate pervasive science research with social and environmental science. It helps us to understand city dynamics on a large scale. Various computational tasks can be conducted to analyse an urban social event like crime in low spatio-temporal granularity. In each chapter of this thesis, we focus on modelling mobility dynamics for the predictive surveillance of crime in a real-world scenario. Specifically, we provide solutions for crime event prediction and patrol route planning.Crime event prediction aims to predict the crime event in a region in the near future. In this thesis, we look at dynamic factors that can be used to build a prediction model. Human mobility describes dynamic opportunistic factors that influence crime events. Therefore, coupling human mobility analytics with crime event provides better safety solutions for an urban community. We analyse human mobility data inferred from the location-based social network, foursquare in crime event prediction model. Human mobility helps us to understand the daily activity and location of people in a timely manner. We study relationships between dynamic mobility features and crime patterns. However, urban data collected from heterogeneous sources are large-scale, imbalanced, and sparse in nature. We tackled these issues during the application of the machine-learning algorithm for prediction. In addition to crime event prediction, handling these issues successfully is important to utilise urban data with small spatio-temporal granularity. We verify the proposed dynamic features with several existing ones, including historical, demographic, and geographic features. With general-purpose dynamic features, we also analyse mobility associated crime-specific dynamic features for better prediction result. The visiting profile of an individual user is extracted from historical location-based social networks to estimate the risk associated with that person. This type of risk profiling helps us to understand how many opportunities a person may bring to a location for crime-event occurrences. To utilise such information in prediction tasks, we introduce risk factors for each user and derive user associated crime risk features from these factors. Furthermore, we characterise the crime risk of a region by combining historical crime event and human movements across regions. We derive several region risk associated features to measure the crime risk of a place during a certain period of a day. The user and region risk associated with crime features help to lift the prediction performance of a place during a time interval. The same type of user and region profiling can be applied to other tasks in spatio-temporal domains, which are associated with human mobility.In this thesis, we address the key issues related to another computational task, planning crime patrol routes. Since the surrounding environment is generally uncertain and police have to respond based on the dynamic emergency calls, we propose an optimisation algorithm-based solution to design a police patrol path in the dynamic environment. The proposed algorithm is sensitive to the environment and changes its pre-planned route, based on the emergencies. We also investigate the effect of crime hotspot prediction in patrol route design and observe that it facilitates a police officer more idle time with no compromise in the efficiency of catching crime events. To make the dynamic patrol route planning more suitable for practical applications, we extend the problem formulation for multiple police officers. We apply metaheuristic optimisation algorithms to find the optimal solution. Same as the previous approach, a prediction based method is applied for the representation of possible initial solutions.In summary, this thesis contains efficient techniques that can be used in the analysis of crime event surveillance. Proposed solutions for crime event prediction and patrol route planning can be applied to combat other social science problems from spatio-temporal aspects. This research takes many inspirations from social science research to provide computational solutions for crime event surveillance in real-world scenario. The contribution of this research enables dynamic decision support to improve the city, urban, and regional monitoring.
@phdthesis{RUMIShakilaKhan2020Mdou,
title = {Modelling dynamics of urban mobility for predictive surveillance of crime},
abstract = {The study of urban dynamics is to understand and analyse the dynamic properties of a city in the spatio-temporal domain. It includes the daily routines of the inhabitants, their movement patterns, geography, demography, and the social and economic condition of a city. Such data has created an excellent opportunity to integrate pervasive science research with social and environmental science. It helps us to understand city dynamics on a large scale. Various computational tasks can be conducted to analyse an urban social event like crime in low spatio-temporal granularity. In each chapter of this thesis, we focus on modelling mobility dynamics for the predictive surveillance of crime in a real-world scenario. Specifically, we provide solutions for crime event prediction and patrol route planning.Crime event prediction aims to predict the crime event in a region in the near future. In this thesis, we look at dynamic factors that can be used to build a prediction model. Human mobility describes dynamic opportunistic factors that influence crime events. Therefore, coupling human mobility analytics with crime event provides better safety solutions for an urban community. We analyse human mobility data inferred from the location-based social network, foursquare in crime event prediction model. Human mobility helps us to understand the daily activity and location of people in a timely manner. We study relationships between dynamic mobility features and crime patterns. However, urban data collected from heterogeneous sources are large-scale, imbalanced, and sparse in nature. We tackled these issues during the application of the machine-learning algorithm for prediction. In addition to crime event prediction, handling these issues successfully is important to utilise urban data with small spatio-temporal granularity. We verify the proposed dynamic features with several existing ones, including historical, demographic, and geographic features. With general-purpose dynamic features, we also analyse mobility associated crime-specific dynamic features for better prediction result. The visiting profile of an individual user is extracted from historical location-based social networks to estimate the risk associated with that person. This type of risk profiling helps us to understand how many opportunities a person may bring to a location for crime-event occurrences. To utilise such information in prediction tasks, we introduce risk factors for each user and derive user associated crime risk features from these factors. Furthermore, we characterise the crime risk of a region by combining historical crime event and human movements across regions. We derive several region risk associated features to measure the crime risk of a place during a certain period of a day. The user and region risk associated with crime features help to lift the prediction performance of a place during a time interval. The same type of user and region profiling can be applied to other tasks in spatio-temporal domains, which are associated with human mobility.In this thesis, we address the key issues related to another computational task, planning crime patrol routes. Since the surrounding environment is generally uncertain and police have to respond based on the dynamic emergency calls, we propose an optimisation algorithm-based solution to design a police patrol path in the dynamic environment. The proposed algorithm is sensitive to the environment and changes its pre-planned route, based on the emergencies. We also investigate the effect of crime hotspot prediction in patrol route design and observe that it facilitates a police officer more idle time with no compromise in the efficiency of catching crime events. To make the dynamic patrol route planning more suitable for practical applications, we extend the problem formulation for multiple police officers. We apply metaheuristic optimisation algorithms to find the optimal solution. Same as the previous approach, a prediction based method is applied for the representation of possible initial solutions.In summary, this thesis contains efficient techniques that can be used in the analysis of crime event surveillance. Proposed solutions for crime event prediction and patrol route planning can be applied to combat other social science problems from spatio-temporal aspects. This research takes many inspirations from social science research to provide computational solutions for crime event surveillance in real-world scenario. The contribution of this research enables dynamic decision support to improve the city, urban, and regional monitoring.},
author = {RUMI, Shakila Khan},
keywords = {Human Mobility;Prediction Modelling;Optimization;Crime;Data Mining},
school = {RMIT University},
year = {2020},
}
© 2021 Flora Salim - CRUISE Research Group.