Doctoral Degrees (Military History)

Permanent URI for this collection

https://scholar.sun.ac.za/handle/10019.1/15143

Browse

Browsing Doctoral Degrees (Military History) by Subject "Data mining"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    Development of a supervised machine learning model to enhance urban water system management: a case study of Stellenbosch municipality
    (Stellenbosch : Stellenbosch University, 2023-12) Van der Walt, Rejoice; Theletsane, Kula Ishmael; Stellenbosch University. Faculty of Military Science. School for Security and Africa Studies: Military History.
    ENGLISH ABSTRACT: Globally, the challenges of conserving freshwater resources are becoming increasingly complex. Among the reasons cited by several researchers are the continuing growth of the world’s population, urbanisation, and the adverse effects of climate change on rainfall amounts and cycles. The complexity stems from the fact that human and natural systems are inextricably linked and interdependent. This makes managing urban water systems a major challenge that requires an integrated management approach capable of addressing the increasing variables that are interdependent and interrelated in an urban water system. To this end, tools continue to be developed to assist water resources managers to improve their management strategies, while data-driven methods are currently gaining popularity. Researchers have consistently emphasised the importance of accurately predicting the water demands of an urban water system as a prerequisite for effective freshwater management. However, the increasingly interconnected and interdependent variables that result from the interactions between human and natural systems pose a significant challenge to accurately predicting water demand. Consequently, traditional modelling tools are also increasingly becoming inadequate. The impacts of climate change, which lead to uncertainties in precipitation cycles, and rapid urbanisation are the main causes of the inadequacy of traditional modelling tools, as they cannot accurately quantify the uncertainties that arise in the system. As a result, data-driven machine learning techniques are becoming more common and are currently widely used in the Global North. In contrast, their use in the Global South is currently very limited, which is also true in South Africa. Another challenge posed by climate change is the changes in evapotranspiration and precipitation that limit terrestrial water storage and necessitate the search for alternative water sources. Among several options for alternative water sources, the case study area (Stellenbosch Municipality) has considered the reuse of municipal wastewater. However, to date, Stellenbosch Municipality has not developed this resource to any significant extent. It is therefore imperative to investigate the barriers to the development of this resource in the Stellenbosch Municipality. The main goal of this study was to use technology to develop a strategy for the sustainable management of Stellenbosch Municipality’s urban water system. The transdisciplinary research approach was the overarching research methodology used in this study because it provided the researcher with the flexibility to choose methods from different research traditions. Other research methods used in the transdisciplinary approach included a critical systematic literature review, interactive management, simulation, a standard cross-industry data-mining research process, and a case study. The mixed-methods exploratory sequential research design, characterised by two phases, was applied to the Stellenbosch Municipality as the case study, where the unit of analysis was urban water demand. The first phase consisted of collecting qualitative data through a soft management systems interactive research method from a purposively selected focus group of municipal wastewater specialists and community representatives. The collected qualitative data were modelled using Concept Star decision-making tools. The second phase consisted of quantitative data collection and simulation guided by standard cross-industry processes for data-mining research. Both traditional time series models and supervised machine learning models were developed for forecasting and predicting run-of-river abstraction for the Stellenbosch Municipality. Qualitative studies conducted on the factors that hinder the implementation of municipal wastewater reuse as an alternative water source in the Stellenbosch Municipality found that social issues were the main cause, followed by deficiencies in water laws, policies, and guidelines for the implementation of municipal wastewater reuse projects. The four principles of human-centred design were identified as an appropriate methodology for desirable implementation of wastewater reuse projects in the Stellenbosch Municipality. Quantitative studies that predicted urban water demand in the Stellenbosch Municipality showed nonlinearity between total water consumption and population/household growth, which should be the norm. From the exploratory data analysis (EDA), the variable run-of-river abstraction was set as the dependent variable for the modelling processes. The following models were developed: traditional Autoregressive Integrated Moving Average and Seasonal Autoregressive Integrated Moving Average models and supervised machine learning models; thus AdaBoost, Gradient Boosting, Stochastic Gradient Boosting, Random Forest, and Artificial Neural Networks. The model with the best performance was Random Forest, followed by Stochastic Gradient Boosting, both of which the researcher saved and recommended for production. The study’s application of the transdisciplinary research methodology is a unique contribution to urban water management research. In addition, this study helps to highlight the importance of a human-centred design approach and the use of datadriven supervised machine learning techniques in the management of urban water systems, which the researcher considers a human-centred data-driven technological triad for the management of urban water systems. It is an effective framework for deploying novel approaches to water management in an urban setting that can be applied to other communities.