Table 1 Summary of the best ELM models trained on the four datasets.

From: Multiple similarly effective solutions exist for biomedical feature selection and classification problems

Dataset

ALL2

CNS

Adeno

DLBCL

ELMs in Top20Features

3

1673

10000

859

ELMs in AllFeatures

3

313

9994

174

MinAcc(McTwo)

0.651

0.643

0.878

0.914

MaxAcc(McTwo)

0.837

0.843

0.919

0.987

Top20Features: MinAcc(ELM)

0.833

0.944

1.000

1.000

Top20Features: MaxAcc(ELM)

0.833

1.000

1.000

1.000

AllFeatures: MinAcc(ELM)

0.833

0.889

1.000

0.958

AllFeatures: MaxAcc(ELM)

0.833

1.000

1.000

1.000

  1. The models with accuracies larger than 0.800 were collected for the two difficult datasets ALL2 and CNS, and the accuracy cutoff 0.900 was used for the two easy datasets Adeno and DLBCL. Except the heading row, the first two rows are the number of ELM models using the training matrices Top20Features and AllFeatures, respectively. The next two rows “MinAcc(McTwo)” and “MaxAcc(McTwo)” gave the minimum and maximum binary classification accuracies of the models generated on the same datasets9. The minimum and maximum accuracies of the ELM models with the accuracies larger than the cutoff were listed in the last four rows.
Back to article page