Agent-based distributed planning and coordination has shown promising results in controlling operations in complex systems such as those present at airports. Distributed planning differs from centralised approaches because it is performed by several agents, which coordinate plans with each other in order to meet a global objective. In this research, we examine airport surface movement operations and focus specifically on improving the conflict detection abilities of a Multi-Agent Path Finding (MAPF) approach for distributed planning and coordination using machine learning. Our MAPF proposal is built on top of a distributed CBS-based algorithm implemented in an existing Multi-Agent System (MAS) model of Amsterdam Schiphol Airport (AAS). In the proposed approach, we use a delegated Multi-Agent System (dMAS), firstly, to propagate information related to the intended aircraft paths and, secondly, to perform the conflict detection task of the CBS algorithm. To achieve these, the dMAS accesses a set of Artificial Neural Networks (ANNs), each allocated to specific taxiway segments to obtain traversal time estimates of aircraft intending to use those segments. Propagated aircraft intentions are used as predictors for future traversal time predictions either during the intention propagation phase or during CBS conflict detection. The proposed planning and coordination model and three of its variants were tested on a real-world flight schedule extracted from ADS-B ground tracks. Comparisons with a baseline approach that implements a forward kinematic simulation for conflict detection revealed that dMAS-CBS offered more precise conflict predictions while being less prone to Type I errors. More specifically, under scenarios where the airport operates at peak capacity, dMAS-CBS was twice as precise and produced up to five times fewer false positives predictions than the baseline approach.