An experiential learning-based transit route choice model using large-scale smart-card data

Taking learning into account when modelling passengers’ route choice behaviour improves understanding and forecasting of their preferences, which helps stakeholders better design public transport systems to meet user needs. Most empirical studies have neglected the relationship between current choices and passengers’ past experiences that lead to a learning process about route attributes. This study addresses this gap by using real observed choices from smart-card data to implement a route choice model that takes into account the learning process of passengers during the inauguration of a new metro line in Santiago, Chile. An instance-based learning (IBL) model is used to represent individually perceived in-vehicle travel time in the route choice model. It accounts for recency and reinforcement of experience using the power law of forgetting. The empirical evaluation uses 8 weeks of smart-card data after the introduction of the metro line. Model parameters are evaluated, and the fit and behavioural coherence achieved by the IBL route choice model is measured against a baseline model. The baseline model neglects passenger learning from experience and assumes that all passengers use only trip descriptive information in their decision-making process. The IBL route choice model outperforms the baseline model from the fourth week after the introduction of the metro line. This empirical evidence supports the notion that after the introduction of a new metro line, passengers initially rely on descriptive travel information to estimate travel times for new alternatives. After a few weeks, they begin to incorporate their own experiences to update their perceptions.