Skip to content

Instantly share code, notes, and snippets.

@Steppschuh
Last active December 20, 2016 14:15
Show Gist options
  • Save Steppschuh/0e79c87f49eae49ad170f3851fe3f7f9 to your computer and use it in GitHub Desktop.
Save Steppschuh/0e79c87f49eae49ad170f3851fe3f7f9 to your computer and use it in GitHub Desktop.

Comparing Classifications

Testing Walking classifier using data from records that matched pattern:

recorded_records.*phone.*walking_right-back

Using 4 features for specific classifications:

  • Standard Deviation Accelerometer
  • Variance Accelerometer
  • Mean Absolute Difference Accelerometer
  • Mean Accelerometer

Training for Thomas Kellermeier

Testing against data from Thomas Kellermeier

210 classifications calculated, final result is 90%

Training ratio 5% 10% 15% 20% 25% 50% 75% 100%
Classification 54% 72% 69% 74% 85% 84% 91% 90%
Delta to final -36% -19% -21% -16% -5% -7% 0% πŸ”“

Testing against data from Franziska Eder

124 classifications calculated, final result is 14%

Training ratio 5% 10% 15% 20% 25% 50% 75% 100%
Classification 6% 13% 11% 12% 15% 16% 14% 14%
Delta to final -9% -2% -3% -2% 1% 2% 0% πŸ”’

Testing against data from Stephan Schultz

108 classifications calculated, final result is 12%

Training ratio 5% 10% 15% 20% 25% 50% 75% 100%
Classification 2% 17% 11% 12% 13% 12% 12% 12%
Delta to final -9% 5% -1% 0% 1% 1% 0% πŸ”’

Training for Franziska Eder

Testing against data from Thomas Kellermeier

210 classifications calculated, final result is 80%

❗ Classification is higher than expected, this would be a false positive

Training ratio 5% 10% 15% 20% 25% 50% 75% 100%
Classification 9% 38% 26% 64% 63% 64% 57% 80%
Delta to final -71% -42% -54% -17% -17% -16% -23% πŸ”“

Testing against data from Franziska Eder

124 classifications calculated, final result is 85%

Training ratio 5% 10% 15% 20% 25% 50% 75% 100%
Classification 46% 49% 45% 65% 75% 79% 81% 85%
Delta to final -39% -36% -40% -20% -10% -6% -4% πŸ”“

Testing against data from Stephan Schultz

108 classifications calculated, final result is 53%

Training ratio 5% 10% 15% 20% 25% 50% 75% 100%
Classification 22% 20% 16% 37% 45% 52% 39% 53%
Delta to final -31% -33% -37% -15% -8% -1% -13% πŸ”’

Training for Stephan Schultz

Testing against data from Thomas Kellermeier

210 classifications calculated, final result is 42%

Training ratio 5% 10% 15% 20% 25% 50% 75% 100%
Classification 75% 69% 64% 54% 53% 61% 43% 42%
Delta to final 33% 26% 22% 11% 11% 19% 0% πŸ”’

Testing against data from Franziska Eder

124 classifications calculated, final result is 47%

Training ratio 5% 10% 15% 20% 25% 50% 75% 100%
Classification 73% 36% 26% 23% 24% 37% 36% 47%
Delta to final 26% -11% -21% -24% -23% -10% -11% πŸ”’

Testing against data from Stephan Schultz

108 classifications calculated, final result is 86%

Training ratio 5% 10% 15% 20% 25% 50% 75% 100%
Classification 98% 80% 82% 78% 80% 85% 84% 86%
Delta to final 12% -6% -4% -8% -6% -1% -2% πŸ”“
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment