Show simple item record

dc.contributor.authorYao, An
dc.date.accessioned2024-12-10T12:28:57Z
dc.date.available2024-12-10T12:28:57Z
dc.date.issued2024
dc.identifier.urihttps://hdl.handle.net/11250/3169082
dc.descriptionMasteroppgave(MSc) in Master of Science in Business Analytics, Handelshøyskolen BI, 2024en_US
dc.description.abstractThe emergence of bike-sharing systems (BSS) marks a significant shift in ur-ban mobility, offering sustainable solutions to alleviate traffic congestion, reduce emissions, and enhance transportation access. By 2023, over 800 cities worldwide have adopted these systems, highlighting their rapid expansion and challenges. This thesis focuses on the Oslo City Bike system, emphasizing the importance of accurate demand forecasting for service excellence and user satisfaction. Advanced methodologies predicted hourly station-level bike demand in Oslo, incorporating diverse spatial and temporal attributes. Four predictive mod-els—Gated Recurrent Unit (GRU), Long Short Term Memory (LSTM), GCN-LSTM, and A3T-GCN—were compared. The LSTM model achieved superior prediction accuracy, especially with comprehensive external features and ex-tended training periods. Despite the inferior performance of GCN-based models, valuable insights were gained regarding network connection optimization and geographical considerations. Key trade-offs include balancing model accuracy and complexity, and the impact of extended training periods. The GCN-LSTM model captured spatial-temporal dynamics without external features, offering a flexible demand forecast-ing approach. Challenges for GCN-RNN models, such as sensitivity to features and computational constraints, were also highlighted. Future research should focus on automated edge construction, integrating attention mechanisms with GCN and LSTM, and expanding the model’s ap-plication to the entire network. These advancements could enhance forecasting precision and deepen understanding of bike-sharing usage drivers. In conclusion, this thesis underscores the potential of LSTM and GCN-LSTM models for urban bike-sharing systems, highlighting opportunities for refinement and increased contextual awareness. The findings offer practical implications for improving operational efficiency and customer satisfaction in urban mobility systems.en_US
dc.language.isoengen_US
dc.publisherHandelshøyskolen BIen_US
dc.subjectbusiness analyticsen_US
dc.titlePredictive Modeling for Urban Bike-Sharing Systems: A Comparative Analysis of LSTM and GCN-Based Modelsen_US
dc.typeMaster thesisen_US


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record