Classify type of motion using decision trees and features

Question

Consider a decision tree for separating between the following activities: Sitting, Driving, Running, Walking. Assume that you are given the following features: Dominant frequency (F1), meanX, meanY, and meanZ (mean acceleration along the three axes).

I know that we can separate walking from running using the Dominant Frequency Parameter. I also know that I can separate Driving from Sitting by looking at the mean for x if that is the horizontal axis. How do I go about separating Driving from (Walking and Running)?

Answer 1

You don't need fancy signal processing at this stage. I would attack this as an exploratory data analysis problem. Make a pairwise scatter plots of all these features for different activities color coded.

Eg. in python you can use seaborn.pairplot with hue='activity' option.

If your features are adequate, you will be able visualize clusters for different activities in the feature space. If you don't see clustering then you have a signal processing problem of extracting relevant features that can distinguish these activities better.