For a quick overview of the paper, please refer to the slides

TrajectoryNet: An Embedded GPS Trajectory Representation for Point-based Classification Using Recurrent Neural Networks

Authors: Xiang Jiang, Erico N de Souza, Ahmad Pesaranghader, Baifan Hu, Daniel L. Silver and Stan Matwin

Abstract: Understanding and discovering knowledge from GPS (Global Positioning System) traces of human activities is an essential topic in mobility-based urban computing. We propose TrajectoryNet-a neural network architecture for point-based trajectory classification to infer real world human transportation modes from GPS traces. To overcome the challenge of capturing the underlying latent factors in the low-dimensional and heterogeneous feature space imposed by GPS data, we develop a novel representation that embeds the original feature space into another space that can be understood as a form of basis expansion. We also enrich the feature space via segment-based information and use Maxout activations to improve the predictive power of Recurrent Neural Networks (RNNs). We achieve over 98% classification accuracy when detecting four types of transportation modes, outperforming existing models without additional sensory data or location-based prior knowledge.

Contact

[email protected]

Please feel free to email me in case you have any questions with the code. There are a lot of redundant codes that I have not got a chance to clean up. Sorry about this clutter!

The follwoing softwares are required for this source code.

• Python 3.5
• Tensorflow 1.4.1
• sklean 0.18.1
• numpy 1.12.0
• R (optional for preprocessing)

The original dataset can be found here. We preprocessed a subset of the data in the data directory. This dataset contains features in .csv format as well as .npy format:

• csvdata.csv: include data of four classes (car, bike, drive and walk) with longitudes, latitudes, speeds and other features
• encodedcsv.zip: the same data after discretization and one-hot-encoding
• npy.zip: the same data in .npy format. This can be used to train the model directly.

If you want to train the model without any preprocessing, please unzip the numpy data before training the model:

cd ./data
unzip npy.zip

The file config.json includes the configurations of the model:

• training, validation and test set selection
• number of hidden nodes in each layer
• learning rate
• mini-batch size
• number of layers
• number of epochs
• whether to checkpoint the model after training
• whether to restore the model before training
• the size of truncated sequences
• frequency to evaluate the model while training
• number of threads
• whether to use GPU during training
• etc.

python trajectoryNet.py

In case you are interested in the preprocessing and discretization of the data, please refer to file preprocess.R.

After the preprocess data are stored in a .csv file, it is required to run create_npy.py to transform the data into .npy format to get ready to import to Tensorflow.