Dataset Messidor retinopathy

Basic characteristics Messidor retinopathy

12352

instances

1200

bags

687

features

654

positive bags

between 8 and 12

instances per bag

546

negative bags

Results Messidor retinopathy

AUC dataset 333
classifier error
AUC EER meanPrec avCost
APR - - - -
DiverseDensity (10.000000) - - - -
EMDD (10.000000) - - - -
MILBoost (100 rounds) 73.0 (2.5) 32.7 (1.8) 79.0 (2.3) 18.9 (0.8)
Citation 1NN, c=1 61.5 (0.5) 40.2 (0.7) 62.7 (0.3) 23.3 (0.1)
Citation 3NN, c=5 66.5 (0.4) 37.7 (0.9) 70.7 (0.3) 21.5 (0.1)
MI-SVM p=1.000000 76.1 (0.3) 30.7 (0.2) 79.5 (0.3) 17.9 (0.2)
MI-SVM r=10.000000 71.9 (0.3) 35.6 (0.4) 77.9 (0.3) 19.1 (0.1)
MILES p=1.000000 71.0 (1.0) 34.4 (0.8) 77.0 (1.1) 19.6 (0.3)
MILES r=10.000000 70.3 (0.3) 35.0 (0.5) 75.3 (0.5) 19.9 (0.3)
SVmil mahal.K - - - -
SVmil haussd.K 48.2 (12.3) 51.0 (8.3) 61.1 (6.6) 23.5 (1.4)
SVmil emd.K 47.4 (14.7) 51.5 (10.2) 61.0 (7.9) 23.5 (1.7)
SimpleMIL with Logistic2 76.7 (0.4) 29.7 (0.8) 80.2 (0.6) 17.7 (0.3)
mean-inst+Logistic2 69.1 (0.6) 32.5 (1.1) 69.2 (0.5) 20.8 (0.3)
extremes+Logistic2 69.6 (0.6) 33.8 (0.9) 70.4 (0.5) 20.6 (0.2)
cov-coef+Logistic2 73.7 (0.7) 30.8 (0.8) 76.2 (0.5) 18.8 (0.3)
Bag of Words (k=10)+Logistic2 57.1 (0.9) 45.5 (0.9) 62.3 (0.8) 23.2 (0.2)
Bag of Words (k=100)+Logistic2 63.3 (1.2) 40.6 (1.1) 69.7 (0.8) 21.8 (0.2)
PposteriorMIL 50.0 (0.0) 50.0 (0.0) 54.5 (0.0) 25.0 (0.0)
mean-inst+LIBSVM - - - -
extremes+LIBSVM - - - -
cov-coef+LIBSVM - - - -
Bag of Words (k=10)+LIBSVM - - - -
Bag of Words (k=100)+LIBSVM - - - -

Results are obtained using 10-fold stratified crossvalidation. The best results are indicated in bold, together with the results that are not significantly worse (based on a paired-difference t-test with a confidence level of 0.05). The standard deviations are shown between brackets.

Dissim.repr. AUC dataset 333
classifier error
AUC EER meanPrec avCost
minmin.K+Logistic2 66.6 (0.0) 35.6 (0.0) 67.7 (0.0) 21.3 (0.0)
summin.K+Logistic2 - - - -
meanmin.K+Logistic2 - - - -
meanmean.K+Logistic2 - - - -
mahal.K+Logistic2 - - - -
haussd.K+Logistic2 - - - -
emd.K+Logistic2 - - - -
linass.K+Logistic2 - - - -

Results are obtained using 10-fold stratified crossvalidation. The best results are indicated in bold, together with the results that are not significantly worse (based on a paired-difference t-test with a confidence level of 0.05). The standard deviations are shown between brackets.