Calender
05
June
x-jobb Ellinor Augustsson
Ellinor Augustsson presents her master’s thesis with title: Comparative Study Between Machine Learning and Statistical Methods for Predicting Travel Patterns from Ticket Validation Data
Abstract:
This thesis investigates the potential of machine learning (ML) models to forecast public transport demand patterns, using historical ticket validation data from the Skånetrafiken system in southern Sweden. Focusing on the corridor between Malmö and Lund from 2018 to 2021, the study explores whether ML methods—especially Long Short-Term Memory (LSTM) neural networks—can outperform classical statistical approaches in predicting hourly travel behavior. Multivariate LSTM models incorporating temporal features (hour, weekday, and month) are benchmarked against naive and classical regression models, including Extra Trees Regressors, with evaluation metrics such as RMSE, MAE, and R². The thesis also assesses model performance under stable conditions (Spring 2019) and during external disruptions (Spring 2020, during COVID-19). Results show that both ML and ensemble-based methods outperform naive baselines, with Extra Trees achieving the highest overall accuracy. However, all models struggle to adapt to abrupt behavioral shifts, indicating the importance of retraining or incorporating external contextual features. The findings provide insights for improving real-time transit forecasting and enhancing the user experience in applications like Skånetrafiken’s mobile app.
Examiner:
Kalle Åström, Centre for Mathematical Sciences, Lund University
Supervisors:
Alexandros Sopasakis, Centre for Mathematical Sciences, Lund University
Om händelsen
Tid:
2025-06-05 13:15
till
14:15
Plats
MH:309A
Kontakt
alexandros.sopasakis@math.lth.se