Each row represents one window.

Columns:

1) Test:
baseline -> fall detection not based on positioning (Training set with data from different positions)
withALm -> fall detection based on classified position (Training set only from this position)

2) Robustness:
The estimated position is changed after x windows in sequence are classified as the same position

3) WindowsFilter:
true: use only windows where all frames are "FALL" or all frames are "ADL" but not windows with mixed frames
false: use all windows
xxx: use all windows

4) Dataset:
(2=MMSys, 3=UMA)

5) Subject ID:
--> 2XXX Dateset ID (2=MMSys, 3=UMA, 4=MiBShar, 5=SisFall)
--> XX01 Subject number

6) Window ID:
Starting time of the window in milli seconds used as ID

7) RealPosition:
Actual position of the sensor in this window

8) ClassifiedPosition:
Classified position by the random forest classifier for positioning

9) Classifier:

The names stand for the following algorithms used:
ANNRangeAlgorithmGravity -> ANN with feature vector 1
J48FallDetectionAlgorithmGravity -> J48 decision tree with feature vector 2
KNNAlgorithmGravity -> KNN with feature vector 2
RandomForestAlgorithmGravity -> Random Forest with feature vector 2
SisFallAccuracyThresholdAlgorithm -> Threshold algorithm with feature vector 3, optimized for accuracy fall
SVMKauAndChen2014Algorithm -> SVM with feature vector 4

Feature vector 1:
- Acceleration range
- Gravity
Feature vector 2:
- Mean accleration
- Variance in acceleration
- Standard deviation in acceleration
- Median of acceleration
- Interquartile range
- Mean absolute deviation
- Kurtosis
- Correlation coefficient
- Entropy (Time)
- Energy
- Standard deviation magnitude horizontal 
- Sum vector magnitude
- Acceleration range
- Gravity
Feature vector 3:
- Standard deviation magnitude horizontal
Feature vector 4:
- Standard deviation in acceleration
- Sum vectore magnitude


10) Guess:
Classified activity (0=Fall, 1=ADL)

11) Truth:
Actual activity (0=Fall, 1=ADL, 2=UNDEFINED)