Total knee and hip arthroplasty (TKA and THA, respectively) are common and resource intensive procedures contributing to significant burden on healthcare systems. The overall aim of this thesis project is to identify and develop strategies to improve elective surgical scheduling for TKA and THA. Machine learning and optimization to predict resource utilization-related outcomes such as duration of surgery and length of stay for arthroplasty patients were identified. A predict-then-optimize approach utilizing neural network models and linear programming was compared to historic scheduling strategies. Data sources for this work include the administrative National Surgical Quality Improvement Program and an institutional database from the Holland Centre at Sunnybrook Health Sciences Centre. The most important features of the neural network models for outcome prediction were identified and compared between databases. Together, these findings are the foundation upon which to develop a “smart” surgical scheduling system and improve the efficiency of operating room utilization.