A day-to-day travel time anticipation and learning framework under imperfect information

Within rapidly evolving emerging transport systems, travellers operate in dynamic, stimulus-rich environments characterised by greater flexibility and diverse real-time information. Learning and re-evaluation have become central to travel decision-making, with individuals refining expectations based on personal experience and external information such as advanced traveller information systems, traffic signage, and social interactions. However, learning is neither uniform nor unimpeded, as cognitive constraints, selective attention, and habitual effects influence belief updating. These factors underscore the need for behaviourally grounded models that account for varying learning profiles and inertia effects. Departing from the Rational Expectations paradigm, this study develops an Attention-Based Travel Time Learning framework integrating a continuous quadratic Rational Inattention model for within-day travel time anticipation with a non-parametric kernel density estimation mechanism for day-to-day belief updating. The resulting multitemporal framework captures how boundedly rational travellers selectively process information under attention costs, refine expectations, and update priors in response to disruptions. The proposed approach is evaluated alongside established day-to-day learning mechanisms, assessing the extent to which it reproduces empirically documented behavioural patterns. The findings underscore the model’s ability to account for habituality and resistance to change beyond full-information neoclassical assumptions: under higher information-processing costs, travellers exhibit greater reliance on the habitual prior, while lower attention costs facilitate quicker integration of new information and more responsive adjustments. Rationally inattentive travellers are found to cluster expectations around distinct “anchor” levels, balancing the trade-off between anticipation accuracy and the perceptual cost of information processing.