Agent-Based Modelling and Machine Learning in Children’s Social Care

White, Luke Andrew (2025) Agent-Based Modelling and Machine Learning in Children’s Social Care. Doctoral thesis, Birmingham City University.

Preview

Text Luke Andrew White PhD Thesis_Final Version_Final Award December 2025.pdf - Accepted Version Download (12MB)

Abstract

Children’s Social Care is a critical service provided by Local Authorities within the UK, where children and their families are provided support through various means including early interventions, child protection and residential care. These services have been under considerable pressure in recent years due to funding changes since 2010 and rising demand. The importance of understanding the best policy and practice for social workers is key to ensure that these services operate, and that they can improve into the future. Data available in this context to use for conducting analysis of existing policy and outcomes is met with difficulties such as Data Sensitivity and Data Quality. Furthermore, due to the context of such data, it is critical that any analysis, modelling, forecasts and predictions conducted on this data be interpretable and scrutinised to ensure that results can be trusted before decisions are taken. Addressing these challenges is the core of this thesis, with three contributions which together allow for analysis of Children’s Social Care data that utilises novel techniques to provide new insights into the policy and practice of Children’s Services within Local Authorities.

Firstly, Agent-Based Simulations using Genetic Algorithm Calibration: This contribution presents the use of Agent Based Models to enable the use of existing policy to inform a model’s design, where a population of Social Workers acting within a Local Authority can be simulated, taking advantage of the limited quality data that exists to configure the model. Further, this approach is presented with a Calibration method that can optimise the model’s parameters to the existing data to validate its design and to provide new insights from the limited data. This approach also benefits from the mitigation of possible risks of using sensitive data directly for modelling, as the model design can be indirectly informed by such data, without training, and any data produced by the model will be synthetic, removing any potential risks.

Second, Temporal Meta-optimiser based Sensitivity Analysis (TMSA) for Agent-Based Models: This contribution outlines a novel approach to Sensitivity Analysis for Agent-Based Models, like the one presented in the first contribution. The use of Sensitivity Analysis in the validation of model design is important in ensuring a model’s creditability and enables the interpretation of model behaviour. With existing methods being ill-suited to Agent Based Models, the TMSA method is presented that utilises novel machine learning approaches to conduct this form of analysis. With TMSA, Agent-Based models can be interpreted and scrutinised more effectively by those designing them.

Finally, Reinforcement Learning based process optimisation for Agent-Based Models: This contribution takes the previous methods developed and utilises them to create a novel approach to optimising Agent-Based Model process design. The approach uses Reinforcement Learning to identify changes in the code of an Agent-Based Model that will lead to an improvement in the model’s ability to represent existing data, through the use of both Sensitivity Analysis and Calibration. The approach further provides better understanding and interpretation of model designs, with an ability to identify shortcomings with assumptions, thus potentially challenging the existing policies and practices of Children’s Social Care.

These contributions together mark a significant step in the application of Agent-Based Models and Machine Learning in the Children’s Social Care context, where they can improve the analytical capabilities of Local Authorities by overcoming the challenges regarding data that they face.

Item Type:	Thesis (Doctoral)
Dates:	Date Event 11 December 2025 Accepted
Uncontrolled Keywords:	Children’s Social Care, Children’s Services, Agent-Based Model, Machine Learning, Genetic Algorithm, Reinforcement Learning, Sensitivity Analysis
Subjects:	CAH11 - computing > CAH11-01 - computing > CAH11-01-01 - computer science CAH11 - computing > CAH11-01 - computing > CAH11-01-05 - artificial intelligence CAH15 - social sciences > CAH15-04 - health and social care > CAH15-04-01 - social work CAH15 - social sciences > CAH15-04 - health and social care > CAH15-04-02 - childhood and youth studies
Divisions:	Architecture, Built Environment, Computing and Engineering > Computer Science Doctoral Research College > Doctoral Theses Collection
Depositing User:	Louise Muldowney
Date Deposited:	16 Dec 2025 09:27
Last Modified:	16 Dec 2025 09:27
URI:	https://www.open-access.bcu.ac.uk/id/eprint/16775

Actions (login required)

View Item