Abstract
The rapid development of technology and telecommunications, particularly in the wake of the COVID-19 pandemic, has transformed healthcare delivery, positioning telemedicine as a vital solution to logistical and geographical barriers. Despite its potential, telemedicine systems face significant challenges, including prolonged patient waiting times, inefficient resource utilisation, low doctor availability (as evident in South Africa), and fluctuating patient loads. This study proposed the application of queuing theory, artificial neural networks (ANN) and a hybrid ANN coupled with particle swarm optimisation (ANN-PSO) to address these operational challenges and enhance telemedicine efficiency.
In this research, the Revoledu online queuing system tool was employed to generate simplified, scenario-based data, which served as input for the development and testing of ANN and hybrid ANN-PSO models. The primary focus of the study was the implementation and evaluation of these models using MATLAB. Both ANN and ANN-PSO models were developed and optimised in MATLAB, with their performance assessed through visualizations such as regression plots and other evaluation metrics.
The study began with a parametric analysis of the queuing system, examining the impact of factors such as server capacity, service rates, and patient arrival rates on system performance. Findings indicated that increasing server capacity or improving service rates significantly reduced wait times and queue lengths. However, these improvements diminished beyond a certain threshold, demonstrating the principle of diminishing returns. Conversely, higher patient arrival rates led to longer queues and greater server congestion, emphasising the importance of aligning system capacity with patient demand trends. This analysis was conducted using 20 simulated scenarios, with input parameters varied individually within defined ranges while others were kept constant. The results underscored the need to balance cost-effectiveness and resource allocation for seamless queueing in telemedicine.
Building on the parametric study, ANN models were created to forecast important performance metrics such as queue intensity, server utilisation, queue length, and delays. Using MATLAB, the 100 data points generated through Revoledu were divided
6
into training (70%), validation (15%) and testing (15%) sets. The ANN models demonstrated that they could dynamically predict output metrics with low mean squared error (MSE) and high correlation coefficient values, confirming the models’ reliability and robustness. These results highlighted the utility of ANN in telemedicine for providing actionable insights, enabling administrators and stakeholders to make informed decisions regarding system planning and resource allocation.
To further improve the performance of the telemedicine queuing system, the study applied the ANN-PSO algorithm. This model, also developed in MATLAB using the same 100 scenarios—but this time split into 90% training and 10% testing—demonstrated superior optimisation capabilities. The model provided the best configuration in terms of the number of neurons, acceleration factors, and swarm population size to be considered, offering practical guidance for telemedicine administrators to improve system performance and manage possible fluctuations effectively. This robust model illustrated the potential of combining machine learning with optimisation techniques to solve the dynamic and complex problems that telemedicine systems encounter.
In conclusion, this study underscored the importance of a holistic approach to telemedicine system design, integrating queuing theory with advanced computational models for optimisation. By leveraging tools such as Revoledu and MATLAB, and incorporating ANN and ANN-PSO, this research presented a comprehensive framework for improving patient experiences, optimising resource utilisation, and ensuring scalability in telemedicine queuing systems. Future studies should explore more complex datasets, additional performance indicators, and dynamic scheduling algorithms to further refine telemedicine applications and adapt to diverse and evolving healthcare scenarios.