Synopsis

This repository contains code and jupyter notebooks with machine learning algorithms for working with GPS trajectories. It will be used during the Machine Learning hackathon of IotTechDay2017.

The dataset used is the popular GeoLife GPS Trajectories

Tasks

• 0. How can we load GPS trajectories in a proper way so that it will be easier to work with in the future.
• 1. Supervised Machine Learning: Build a classifier which can automatically detect the transportation mode of the trajectories (walking, bicycle, car etc).
• 2. Unsupervised Learning; Clustering of the GPS trajectories by using auto-encoders and recurrent neural networks.
• 3. GeoSpatial analysis of the GPS trajectories; Analysis and visualization of the taken routes (does the popularity of a route affect the traffic? What are the points of interest e.g., restaurants, stores, hotels, etc. )

Notebooks

We have provided some notebooks, which should give you a flying start, but feel free to do everything your own way.

• 1. loading the GPS data, enriching it with velocity and acc information, adding correct labels to each trajectory

• 2. Modality detection with enriched and labeled trajectories

Possible relevant datasets

• Bus routes and stops of Beijing -> contains 1.543 bus routes and 42.161 stops

• Beijing check-in records from Sina -> contains 868 m check-ins for all 143,576 venues.

• Road junction density of China in 2011

• Points of interest in Beijing

• Global cities shapefile

• Beijing bus smart card data

Interesting articles:

• China's GPS Shift problem
• Homemade GPS receiver