Abstract:
The rapid growth of urban populations and increasing traffic congestion have prompted
the need for efficient traffic management systems. Accurate traffic flow forecasting is
crucial for optimizing transportation infrastructure, reducing congestion, and enhancing
safety. This thesis explores the use of deep learning techniques to predict real-time traffic
flow, aiming to develop an intelligent system capable of providing timely and accurate
traffic predictions. The study leverages several deep learning models, including
Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Long
Short-Term Memory (LSTM) networks, to forecast traffic conditions based on historical
traffic data. Through experimentation with multiple datasets, including urban traffic data
and public traffic flow databases, the research investigates the effectiveness of these
models in capturing complex traffic patterns. Key performance metrics such as Mean
Absolute Error (MAE), Root Mean Square Error (RMSE), and R² are used to evaluate the
models’ prediction accuracy. The results demonstrate that deep learning models,
particularly hybrid CNN-LSTM models, outperform traditional forecasting methods,
offering improved accuracy and adaptability in dynamic traffic environments. Challenges
such as data quality, real-time prediction constraints, and the influence of external factors
like weather and events are also addressed. The findings suggest that deep learning has
the potential to revolutionize traffic management by providing more accurate and timely
forecasts, thereby aiding in the development of smarter, more efficient transportation
systems. Future work includes exploring multi-modal data integration and real-time
prediction implementation to further enhance the capabilities of traffic forecasting
systems.
