Abstract
With the advent of Artificial Intelligence, more advanced models have been developed and applied in healthcare services. To aid management and service teams to allocate resources and optimise different processes efficiently is of great importance. Specifically, the field of predicting patient numbers and their trends with AI solutions to improve operational efficiencies and quality of healthcare has a vast untapped potential. In this thesis, machine learning related techniques are provided in the health services applications, especially in emergency department (ED) visits forecasting and in dengue trends forecasting. Fluctuations in ED patient volumes make allocation of resources in a manner that reduces wastage and cost while enabling consistent patient care challenging. Globally, ED crowding and acute hospital bed shortages make these considerations increasingly important. Accurate forecasting of ED flows allows real-time resource planning to optimize manpower, materials and other processes in order to improve cost-effectiveness of hospital management. Dengue trends forecasting is also very important in the local context and many other countries elsewhere, as forecasting of the weekly cases accurately can help health agencies allocate limited resources effectively and form more precise plans for vector control measures. A deep stacked architecture and a deep learning hybrid architecture are introduced and applied to problem statement in the healthcare services. The performance of the stacked architecture based models and hybrid based models are compared with leading models in the literature with our proposed architectures outperforming them. Keywords: Machine learning, Deep learning, Deep stacked architecture, Deep hybrid models, Time series forecasting, ED visits prediction, Dengue cases forecasting