Online stochastic assignment problem with feature-based demand learning

Abstract

This paper focuses on demand learning through the utilisation of unknown features to optimise resource allocations. The performance of Greedy, Simulate-Optimize- Assign-Repeat (SOAR) and Random algorithms are compared with synthetic and real-world private-hire car data. Through the control testing with synthetic data, SOAR outperforms the other algorithms. Additionally, the findings also highlights the impact of input parameter variability on algorithm performance. Furthermore, the examination of real-world data are in consistent with these findings, emphasiz- ing the practical relevance of the study’s outcomes. Acknowledging these variations enables better decision-making, tailoring algorithms to specific complexities, and enhancing overall outcomes.