Dataset taken from http://archive.ics.uci.edu/ml/datasets/Heterogeneity+Activity+Recognition# Be aware that when unzipped, data is ~3.3gb! data not included in repository(just download it from link above)
- download the SensorUDP app from https://play.google.com/store/apps/details?id=com.ubccapstone.sensorUDP&hl=en_US
- run The Live_Sensor_Streaming jupyter notebook
- follow instructions and enjoy! (due to controlled conditions in the dataset, it only works if the back of the phone is facing downward, I try to do some coordinate system transformations using the rotation sensors, but haven't completely nailed it down)
- Get a more representative dataset
- Different train test split (time split/leave one out?)
- Include more activities
- Include more sensors if accelerometer isn’t sufficient
- Include more time and frequency features