Traffic flow prediction using bi-directional gated recurrent unit method.

Area of Science:

• Intelligent Transportation Systems
• Data Science
• Machine Learning

Background:

• Accurate traffic flow prediction is crucial for intelligent transportation systems.
• Traditional models struggle with the complex, non-linear temporal dynamics of traffic data.
• Deep learning models offer potential for enhanced prediction accuracy.

Purpose of the Study:

• To evaluate the effectiveness of a Bi-directional Gated Recurrent Unit (Bi-GRU) model for traffic flow prediction.
• To compare the Bi-GRU model's performance against benchmark models including ARIMA, LSTM, Bi-LSTM, and GRU.
• To analyze the model's ability to capture temporal traffic flow characteristics.

Main Methods:

• Utilized a Bi-directional Gated Recurrent Unit (Bi-GRU) model for traffic flow prediction.
• Applied data preprocessing techniques including Augmented Dickey-Fuller unit root test and differencing.
• Compared Bi-GRU performance against ARIMA, LSTM, Bi-LSTM, and GRU using real-world traffic data.

Main Results:

• Bi-GRU demonstrated superior performance with RMSE of 30.38, MAPE of 9.88%, and MAE of 23.35.
• Deep learning models (LSTM, Bi-LSTM, GRU, Bi-GRU) significantly outperformed the traditional ARIMA model.
• Bi-GRU achieved a 0.48% lower MAPE than GRU, indicating a marginal improvement in prediction error.

Conclusions:

• The Bi-GRU model is highly effective for traffic flow prediction, outperforming other benchmark models.
• Deep learning approaches, particularly Bi-GRU, offer significant advantages over traditional methods for complex traffic data.
• The model shows higher accuracy during peak periods but exhibits a slight lag in prediction.