This is the code for the paper "A Hybrid Method for Intercity Transport Mode Identification Based on Mobility Features and Sequential Relations Mined from Cellular Signaling Data".
To ensure compatibility and reproducibility, the following system configuration was used for development:
- Operating System: Windows 10 CentOS 6.0
- Python Version: 3.8.18
This project depends on specific versions of TensorFlow and scikit-learn:
- TensorFlow Version: 2.4.0
- scikit-learn Version: 1.1.0
To ensure fairness in the experimental evaluation, each model is configured with comparable parameters as outlined below:
| Model | Configuration |
|---|---|
| KNN | 3 neighbors |
| RF | Maximum depth: 10, Estimators: 200, Max features: 10 |
| XGBoost | Estimators: 200, Max depth: 10, Learning rate: 0.01 |
| LSTM | Units: 32, Epochs: 50, Batch size: 72, Optimizer: Adam |
| BiLSTM | Units: 32, Epochs: 50, Batch size: 72, Optimizer: Adam |
| CNN-BiLSTM | Adds convolutional, pooling layer, batch normalization to BiLSTM architecture |
| Hybrid | RFE(Maximum depth: 10, Estimators: 200, Max features: 10), BiLSTM(Units: 32, Epochs: 50, Batch size: 72, Optimizer: Adam) |
The following table summarizes the training and inference times (in seconds) for each model:
| Model | Training Time (s) | Inference Time (s) |
|---|---|---|
| KNN | 0.04 | 0.04 |
| XGBoost | 8.21 | 0.08 |
| LSTM | 45.33 | 2.1 |
| RF | 67.46 | 0.56 |
| Bi-LSTM | 68.34 | 1.54 |
| CNN-BiLSTM | 80.85 | 1.05 |
| Hybrid | 143.60 | 2.01 |
The table below shows the average performance of each model across different evaluation metrics. This summary provides a quick overview of how each model performs in various scenarios, including tolling freeways (FWY), toll-free state highways (HWY), high-speed railways (HSR), normal-speed railways (NSR), and static state (SS), along with the overall average.
| Model | Overall Avg | FWY Avg | HWY Avg | HSR Avg | NSR Avg | SS Avg |
|---|---|---|---|---|---|---|
| KNN | 0.729 | 0.689 | 0.623 | 0.798 | 0.366 | 0.936 |
| XGBoost | 0.891 | 0.916 | 0.763 | 0.890 | 0.832 | 0.960 |
| LSTM | 0.895 | 0.900 | 0.777 | 0.897 | 0.867 | 0.960 |
| RF | 0.885 | 0.912 | 0.754 | 0.875 | 0.819 | 0.961 |
| BiLSTM | 0.890 | 0.896 | 0.769 | 0.887 | 0.860 | 0.958 |
| CNN-BiLSTM | 0.895 | 0.895 | 0.815 | 0.841 | 0.825 | 0.955 |
| Hybrid | 0.920 | 0.937 | 0.801 | 0.930 | 0.926 | 0.956 |