Browsing by Author "Ndibatya, Innocent"

Now showing 1 - 2 of 2
  • Thumbnail Image
    Item
    Agent-based modelling of paratransit as an intelligent complex adaptive system to improve efficiency
    (Stellenbosch : Stellenbosch University, 2021-03) Ndibatya, Innocent; Booysen, M. J.; Stellenbosch University. Faculty of Engineering. Dept. of Electrical and Electronic Engineering.
    ENGLISH ABSTRACT: Urban residents in Sub-Saharan Africa (SSA) face mobility challenges that limit theiraccess to jobs, services, markets, and socioeconomic opportunities. In most SSA cities,public transport is predominantly provided by the inefficient paratransit system – a flex-ible mode of passenger transport consisting of privately-owned, low-capacity unscheduledminibuses and motorcycle taxis. There is growing interest among city authorities andurban transport researchers in addressing the inefficiency problem associated with para-transit. Several approaches, such as complete overhaul to bus rapid transit (BRT), andphased banning of paratransit from the cities have previously been proposed and con-comitant implementation projects started. However, most of such projects have eitherfailed to take off, or they have stalled. This is likely because of the huge capital invest-ment required, the unique social and cultural dynamics associated with “third world”countries, and urban sprawl due to poor city planning. This study departs from the com-mon perspective held by several researchers and city authorities who view paratransit as“chaotic”, thus, the justification for its total overhaul and banning. Instead, this studyaims to leverage the beneficial aspects of existing paratransit – such as flexibility, demand-responsiveness and near-ubiquitous coverage – with the elusive objective of achieving amore efficient paratransit state as a result.Through theoretical modelling, field study and experimental approaches, this studyaimed to improve the efficiency of minibus taxis paratransit systems. The theoretical mod-elling work involved modelling paratransit systems as complex adaptive systems (CAS)and developing an agent-based model (ABM) for minibus taxi operations in an organically-evolved paratransit setting. The field study involved an in-depth investigation of minibustaxi operations in Kampala’s paratransit system, and collection and analysis of minibustaxi movement data that was used to validate the agent-based model. The experimen-tal approaches involved three separate simulation experiments, simulating the minibustaxi transportation dynamics with varying levels of agents’ intelligence and situational awareness. Machine learning methods, such as random forests and convolutional neuralnetworks were used to train agents in the subsequent simulation experiment to improvetheir intelligence during decision making. At each stage, several efficiency metrics’ valuessuch as passenger waiting time and minibus taxi occupancy were collected. The resultsfrom the experiments showed that there was an improvement in the overall efficiency ofthe minibus taxi paratransit system. For instance, the average passenger waiting time re-duced from 1.2 hours to 30 minutes, indicating a 55% improvement. Whereas the averageminibus taxi occupancy increased from 42% to 51%, indicating a 21% improvement. Ac-cordingly, we concluded that improving the micro-level agents’ intelligence and situationalawareness, results in an overall increase in the efficiency of the paratransit system.To the transportation researchers, we recommend further work on using ABM toinclude other modes of paratransit transport such as the three-wheeled rickshaws andmotorcycle taxis (boda bodas). To the city authorities, we recommend the integration ofsmart mobility and ICT applications into the paratransit ecosystem to support journeyplanning, booking, scheduling, and fare collection.
  • Thumbnail Image
    Item
    Modelling of inter-stop minibus taxi movements : using machine learning and network theory
    (2014-12) Ndibatya, Innocent; Booysen, Marthinus J.
    ENGLISH ABSTRACT: Minibus taxis provide affordable alternative transport for the majority of urban working population in Sub- Saharan Africa. Often, these taxis do not follow predefined routes in their endeavours to look for passengers. Frequently, they stop by roadsides to pick up passengers and sometimes go off the main route in an attempt to fill the taxi with passengers to make the trip profitable. In addition, the destinations are changed from time to time depending on the driver. This uncoordinated movement creates a web of confusion to would-be passengers. The key aspects that are not clear to the passengers include; where to get a taxi, the waiting time and the travel time to the destination. These conditions leave taxi passengers at a very big disadvantage. In this research, we applied the concepts of machine learning and network theory to model the movements of taxis between stops. The model can be used to compute the waiting times at the stops and the travel times to a specified destination. Twelve minibus taxis were tracked for 6 months. Density-based clustering was used to discover the formal and informal taxi stops, which were modelled into a flow network with the significant stops as nodes and the frequency of departures between nodes as edges representing the strength of connectivity. A data driven model was developed. From the model, we can predict the time a passenger will have to wait at a stop in order to get a taxi and the trip duration.