This paper proposes an architecture for predictive path planning applied to transportation problems for simulated robotics. The aim is to evaluate scenarios where Machine Learning (ML) strategies contribute to optimizing the path planning in environments modeled as multi-layered flat networks and formulated as Linear Programming (LP) problems. This proposal reduces the computational time  to obtain large logical trajectories in navigation maps queried  by mobile robots to reach targets at minimum cost. In the proposed architecture a dataset with feasible solutions is produced  to represent the optimization of link costs, and the ML strategies use this dataset to discover alternative paths on demand in reduced time. In mobile robotics, multi-layer networks for path planning have importance for complex dynamic systems, and the modeling is generally done with neural networks and deep reinforcement learning. In this paper, the focus is on the evaluation of ML algorithms for the discovery of routes for simulated robotics with the extraction of features from logical links. The statistical results reveal a gain in performance in predicting the optimum solution for a large supervised dataset.