That is the challenge Aparajita Saha explored during her Master’s thesis in Computer Science at Universiteit Leiden, conducted in collaboration with Zight.

In many public transport networks, passengers check in when boarding, but do not check out when leaving the vehicle. This creates an important blind spot: operators know where a journey starts, but not where it ends. And without that destination data, it becomes much harder to determine vehicle occupancy and understand passenger flows.

Aparajita investigated whether transfer learning can help solve this problem.

Her approach combines passenger check-in data with network structures and Points of Interest around stops. Using a Graph Neural Network, the model learns travel patterns from data-rich public transport regions and transfers that knowledge to networks where little or no check-out data is available.

The results are promising. Across the source regions, the model predicted the exact destination stop correctly for roughly half of the boardings. By using the probabilities of different predicted destinations, the model was also able to reconstruct vehicle occupancy more accurately, reducing MAE by approximately 35–45% compared with using only the single most likely destination.

And for us, the research does not end with the thesis.

The methodology and insights from Aparajita’s research can be translated into a practical solution at Zight. This means that public transport operators and authorities dealing with incomplete check-out data can use this approach through Zight to gain better insight into destinations, passenger flows and vehicle occupancy.

A great example of how academic research, data science and real-world public transport challenges can come together.

A big thank you to Aparajita Saha for her work and contribution to Zight, and congratulations on completing your thesis!