Martensitic phase transformations and twinning are key mechanisms for accommodating plasticity in a wide range of structural alloy systems. As deformation progresses, transformed domains (TD) alter the microstructure and form networks with remarkable complexity. The topology of these networks and their evolution is expected to dictate the mechanical response of the host microstructure. To date though, our understanding of TD network evolution is limited due to the multiple interrelated mechanisms involved in network dynamics such as TD nucleation, transmission or lack thereof at grain boundaries, short-range interactions within grains, and incidental interactions across grain boundaries. The development of mechanistic models to predict network evolution and rationalize the relative contributions of different mechanisms is a daunting task requiring both extensive data and advanced models. To address this challenge, in this study we introduce a machine learning framework to predict the outcome to TD-grain boundary interactions using multilayer perceptrons. Applying this approach to the case of twinning in Ti, we then leverage the trained model to virtually generate fingerprints of TD networks and explore the role of the aforementioned mechanisms on the evolution of the network structure. The model predicts contact formation with high accuracy and reveals that while the geometric alignment between TDs is a key factor for cross-grain contact formation, the role of internal stresses must be included to accurately capture contact geometries in poorly aligned configurations. Further, by applying the model to the evolution of virtual domain networks, we demonstrate, for the microstructure studied, that incidental contacts between different TD domains play a major role in dictating the network morphology.