Long Short-Term Memory Models for Improving Route Planning Acceptance

Building routes that are optimal in terms of reducing costs, impact on the environment, and human workload plays a key role in logistics. However, producing efficient routes that satisfy operational constraints does not guarantee that human planners and drivers will actually accept and execute them. It is a long standing problem in optimization: personal knowledge as well as subjective preferences are difficult to fully formalize in a model. Smart city applications sharpen challenges, but open also opportunities, given by the larger data collection possibilities. In this paper, we develop a classification model to recognize routes that meet the preferences of human planners. It exploits data-driven techniques, in a pipeline whose core component is a Bidirectional Long Short-Term Memory Neural Network architecture, which is able to capture the relationships between points visited in sequence in accepted routes. We evaluate our methodology on a real-world case study. Our experiments show that our methods clearly outperform Markovian models from the literature in terms of recognition accuracy, demonstrating their ability to identify longer dependencies in the sequence of points. Integrating our recognition model into optimization algorithms can lead to the generation of efficient routes that also accommodate the implicit preferences of drivers and planners, thus enhancing the acceptance rate of suggested solutions.