# Iterative

## No new features with LDA - iter 0

Training accuracy 0.8951783732448177 +/- 0.012720090197940919
Cross-validation accuracy: 0.6756439070980823 +/- 0.030351682367092743

Fold 0: Training accuracy 0.8764542936288089
Fold 0: Cross-validation accuracy: 0.7361139896373057
Fold 1: Training accuracy 0.9144353899883586
Fold 1: Cross-validation accuracy: 0.6553911205073996
Fold 2: Training accuracy 0.8890097932535365
Fold 2: Cross-validation accuracy: 0.6783852691218131
Fold 3: Training accuracy 0.8938944872554832
Fold 3: Cross-validation accuracy: 0.6725396825396826
Fold 4: Training accuracy 0.9020979020979021
Fold 4: Cross-validation accuracy: 0.6757894736842105

## LDA with finger lengths - iter 1

Training accuracy 0.8923648757021899 +/- 0.015222679837493697
Cross-validation accuracy: 0.6937050858401812 +/- 0.02270823122384891

Fold 0: Training accuracy 0.8742382271468144
Fold 0: Cross-validation accuracy: 0.7253886010362695
Fold 1: Training accuracy 0.9138533178114087
Fold 1: Cross-validation accuracy: 0.6617336152219874
Fold 2: Training accuracy 0.8759521218715995
Fold 2: Cross-validation accuracy: 0.7110481586402266
Fold 3: Training accuracy 0.8962655601659751
Fold 3: Cross-validation accuracy: 0.6924603174603174
Fold 4: Training accuracy 0.9015151515151515
Fold 4: Cross-validation accuracy: 0.6778947368421052

Confusion matrix differences
- Aankomen-a beter 0.04
- C:1 beter 0.03, less confusion with C:2
- GOED-A beter 0.06
- Hebben-A beter 0.04
- C-OOK worse (more confusion with ZELFDE)

## LDA with body distances - iter 2

Training accuracy 0.9026106794835475 +/- 0.010783804739660988
Cross-validation accuracy: 0.6990130512995554 +/- 0.03517041867437573

Fold 0: Training accuracy 0.8880886426592798
Fold 0: Cross-validation accuracy: 0.7564766839378239
Fold 1: Training accuracy 0.9185098952270081
Fold 1: Cross-validation accuracy: 0.6469344608879493
Fold 2: Training accuracy 0.8933623503808488
Fold 2: Cross-validation accuracy: 0.6968838526912181
Fold 3: Training accuracy 0.9057498518079431
Fold 3: Cross-validation accuracy: 0.7063492063492064
Fold 4: Training accuracy 0.9073426573426573
Fold 4: Cross-validation accuracy: 0.6884210526315789

Confusion matrix differences
- ZELFDE-A, C-OOK better but more confused with each other, less confusion with other classes
- C:ZIEN worse because of confusion with C:1
- AANKOMEN improved

## LDA with finger bent angles - iter 3

Training accuracy 0.893942270369088 +/- 0.011037541364725335
Cross-validation accuracy: 0.6909539028661393 +/- 0.02962781166043358

Fold 0: Training accuracy 0.8808864265927978
Fold 0: Cross-validation accuracy: 0.7435233160621761
Fold 1: Training accuracy 0.9109429569266589
Fold 1: Cross-validation accuracy: 0.6638477801268499
Fold 2: Training accuracy 0.8830250272034821
Fold 2: Cross-validation accuracy: 0.6997167138810199
Fold 3: Training accuracy 0.8956727919383521
Fold 3: Cross-validation accuracy: 0.6845238095238095
Fold 4: Training accuracy 0.8991841491841492
Fold 4: Cross-validation accuracy: 0.6631578947368421

Confusion matrix differences:
- HEBBEN-A greatly improved (0.54 -> 0.61)
- C:1 less confused with C:2, more with C:ZIEN

## LDA with trajectories - iter 4

Training accuracy 0.9036142271836509 +/- 0.00929315334754495
Cross-validation accuracy: 0.6914263783673984 +/- 0.033926944497788436

Fold 0: Training accuracy 0.8914127423822714
Fold 0: Cross-validation accuracy: 0.7357512953367875
Fold 1: Training accuracy 0.9161816065192084
Fold 1: Cross-validation accuracy: 0.638477801268499
Fold 2: Training accuracy 0.8944504896626768
Fold 2: Cross-validation accuracy: 0.7110481586402266
Fold 3: Training accuracy 0.9069353882631891
Fold 3: Cross-validation accuracy: 0.7023809523809523
Fold 4: Training accuracy 0.9090909090909091
Fold 4: Cross-validation accuracy: 0.6694736842105263

Confusion matrix differences:
- GOED-A greatly improved

# Combinations

### LDA with finger lengths and body distances - iter 5

Training accuracy 0.9136217662432333 +/- 0.007835002633468478
Cross-validation accuracy: 0.704793349899085 +/- 0.024666139032408798

Fold 0: Training accuracy 0.9041551246537396
Fold 0: Cross-validation accuracy: 0.7357512953367875
Fold 1: Training accuracy 0.9237485448195576
Fold 1: Cross-validation accuracy: 0.6659619450317125
Fold 2: Training accuracy 0.9047878128400435
Fold 2: Cross-validation accuracy: 0.7195467422096318
Fold 3: Training accuracy 0.916419679905157
Fold 3: Cross-validation accuracy: 0.7142857142857143
Fold 4: Training accuracy 0.918997668997669
Fold 4: Cross-validation accuracy: 0.6884210526315789

### LDA with finger bents and body distances - iter 6

Training accuracy 0.9068261372755113 +/- 0.009389008540504756
Cross-validation accuracy: 0.7010384925731634 +/- 0.031657455068212104

Fold 0: Training accuracy 0.9168975069252078
Fold 0: Cross-validation accuracy: 0.7538860103626943
Fold 1: Training accuracy 0.9254947613504074
Fold 1: Cross-validation accuracy: 0.6511627906976745
Fold 2: Training accuracy 0.911316648531012
Fold 2: Cross-validation accuracy: 0.7025495750708215
Fold 3: Training accuracy 0.9264967397747481
Fold 3: Cross-validation accuracy: 0.7023809523809523
Fold 4: Training accuracy 0.9236596736596736
Fold 4: Cross-validation accuracy: 0.6821052631578948