Heuristics are widely used to solve NP-hard combinatorial optimisation problems, but their performance depends heavily on parameter values that are often still chosen manually in Operations Research (OR). This empirical process is costly and difficult to generalise, which explains the growing recent interest in automatic algorithm configuration within OR, inspired by developments in machine learning.

In this work, we apply Bayesian optimisation to tune the parameters of an Ant Colony Optimisation algorithm for a scheduling problem. Our results show that this automatic approach outperforms the best manually tuned configuration reported in the literature. We also provide initial insights into how the algorithm's parameters correlate with each other, suggesting possible structural relationships to explore in future work.