Time Series Gesture Detection
Personal project where I used machine learning to come up with accurate predictions of gestures from multi-channel time-series data. Each of the channels represents the amplitude of a muscle activation in an arm. The data exploration, feature engineering, and feature selection processes can be seen in the Jupyter notebook files, and the final script for training a batch of unclassified gesture files located in a directory is saved.
The final model achieved 86% 70-30 cross-validation accuracy on the dataset.
Libraries
- Pandas
- Numpy
- Scikit-Learn
- TsFresh
- XGBoost
- Pickle
Example
Multi-Channel Data Sample
