Feature Extraction Continued…

We computed a partial list of our features but this needed to be done for you for each of the accelerometer channels in activity_classifier_utils.py.

Summary

Sampling a sensor hundreds of times per second means that raw sensor data has huge dimensionality. There are 7680 points of accelerometer data at 256 Hz over 10 seconds. Trying to model data points of this size will not be successful. We need to do dimensionality reduction.

By selecting features from the literature, we can be confident that we are not using features that overfit to our dataset and that the activity classifier we build will generalize to other studies and devices.

Check out the notebooks for this lesson and activity_classifier_utils.py to see how we implement the features and compute them for our dataset.

Notebook Review

If you wanted to interact with the notebook in the video, you can access it here in the repo /activity-classifier/walkthroughs/feature-extraction-continued/ or in the workspace below.

The dataset that will be used throughout this lesson can be found at the top of the lesson directory at /activity-classifier/data/.

Further Resources

This blog post goes through a very similar process as this lesson. It starts by explaining some signal processing techniques (like we did earlier in the course). The author uses those techniques to build features in much the same way we just did. And then, he uses those features to build an activity classification model, just as we are about to!
Machine Learning with Signal Processing Techniques

Literature

The algorithm we built was inspired by these two papers.

• Mehrang S., Pietilä J., Korhonen I. An Activity Recognition Framework Deploying the Random Forest Classifier and A Single Optical Heart Rate Monitoring and Triaxial Accelerometer Wrist-Band. Sensors. 2018;18:613. doi: 10.3390/s18020613. Link
• Liu S, Gao RX, Freedson PS. Computational methods for estimating energy expenditure in human physical activities. Med Sci Sports Exerc. 2012;44:2138–2146. doi: 10.1249/MSS.0b013e31825e825a. Link