Spatiotemporal Traffic Accident Prediction Using Deep Learning Models

🎓 Master’s Thesis Presentation – Akseli Manninen (Aalto University, 2024)
Title: Spatiotemporal Traffic Accident Prediction Using Deep Learning Models
Supervisor: Prof. Alexander Jung
Advisor: Jun Yang, M.A.

🚧 How can cities predict and prevent traffic accidents before they happen?
This master’s thesis demonstrates how deep learning can support proactive traffic safety planning. Focusing on Helsinki, the study combines historical accident records, traffic volumes, weather, land use, and calendar data to predict weekly accident risk at the neighborhood level.

📊 Key takeaways for city planners:
- Enables targeted interventions in high-risk areas and time windows
- Supports data-driven implementation of Vision Zero strategies
- Informs infrastructure design, enforcement planning, and public outreach
- Leverages freely available urban data sources for practical deployment

🧠 Among the evaluated models—LSTM, CNN-LSTM, and Transformer—a CNN-LSTM hybrid delivered the most accurate predictions, outperforming historical baselines.

🌍 If your city is working to reduce traffic injuries and fatalities, this talk offers practical insights into how AI can enhance urban safety and resilience.

@aaltouniversity #deeplearning #roadsafety @helsinginkaupunki1341 @fintraffic