hki-traffic-predict

Keras neural network to predict traffic in Helsinki

Installation

Note: this project can only be run with Python 3.

pip install -r requirements.txt

Usage

There are several scripts included:

• train_short_term.py - Given the statistics for past 3 hours, make traffic predictions for the next 4 hours
• train_shorter_term.py - Given just the current (now) traffic data, make traffic predictions for the next 4 hours

Check the active_model variable for which model will be used

Then run:

python train_short_term.py

That trains the model and saves it as the name of the script + active_model variable, like short_term_dense_1.h5

Results

short_term

conv1d_1

- 0s - loss: 0.0271 - val_loss: 0.0263

conv1d_2

- 0s - loss: 0.0191 - val_loss: 0.0174

conv1d_3

- 0s - loss: 0.0151 - val_loss: 0.0149

dense_1

- 2s - loss: 0.0330 - val_loss: 0.0296

lstm_1

- 1s - loss: 0.0319 - val_loss: 0.0257

lstm_2

- 7s - loss: 0.0251 - val_loss: 0.0211

lstm_3

- 4s - loss: 0.0278 - val_loss: 0.0240

shorter_term

conv1d_1

1s 78us/step - loss: 0.0261 - val_loss: 0.0231

conv1d_2

1s 82us/step - loss: 0.0236 - val_loss: 0.0205

dense_1

1s 59us/step - loss: 0.0335 - val_loss: 0.0289

dense_2

1s 59us/step - loss: 0.0294 - val_loss: 0.0248

dense_3

1s 63us/step - loss: 0.0343 - val_loss: 0.0302

dense_4

1s 70us/step - loss: 0.0211 - val_loss: 0.0163

lstm_1

1s 78us/step - loss: 0.0239 - val_loss: 0.0207

lstm_2

5s 358us/step - loss: 0.0314 - val_loss: 0.0279

Other

Source of CSV: https://hri.fi/data/dataset/liikennemaarat-helsingissa