Skip to main content

Table 1 Classification accuracy using four multi-class cancer data sets (GCM, Breast, Leukemia, Lymphoma) and six binary sets (CNS, colon, DLBCL, GCM, lung, prostate) show that performance of BCGA-ELM is superior and consistent over all these data sets. GCM multi-class has an accuracy of 95.4%, which is at least 21.6% higher than other methods given in the literature (although some of them use very small sets of genes)

From: Multi-class BCGA-ELM based classifier that identifies biomarkers associated with hallmarks of cancer

Multi – class Binary-class
Data [ 3 , 12 , 20 ] GCM Breast Leukemia Lymphoma CNS Colon DLBCL GCM2 Lung Prostate
#Genes-initial set 16063 1213 999 4026 7129 2000 7129 16063 12533 12600
#Genes BCGA-ELM 92 30 11 27 17 27 18 73 11 72
# Samples 198 49 38 96 34 62 77 280 181 102
# Classes 14 4 3 5 2 2 2 2 2 2
Multi-class, Accuracy (%) Binary-class, Accuracy (%)
BCGA-ELM 95.4 100 100 100 100 100 100 100 100 100
(*σ2 = 0.00083)    
Weka packages [ 19 ]           
LibSVM-linear 78.9 100 100 91.9 100 91.9 100 99.1 95.6 97.1
RBF Network 69.8 100 100 82.3 98.7 79.1 96.2 85.4 96.7 93.6
SMO 83.3 100 100 93.6 98.7 89.7 98.7 98.7 95.0 97.1
Naïve Bayes 78.6 100 97.1 72.6 93.5 60.0 81.9 73.1 97.8 92.7
Multiclass Classifier 85.3 100 100 93.6 97.4 93.5 99.8 99.7 94.5 98.8
Method #Genes  
ICGA-PSO-ELM [ 15 ] 42 88.3 91.2 100 97.0 100 - - - -
HC-k-TSP [8] 5 to 27 67.4 66.7 97.1 - 97.1 90.3 97.4 85.4 97.0
mul-PAM [9] 5 to 27 56.5 93.3 97.1 - 85.3 90.3 92.2 82.9 93.9
BMSF(highest) [10] 5 to 27 - - - - 97.1 95.2 97.4 98.6 100
I-RELIEF(highest)[11] 5 to 27 - - - - 88.4 82.3 95.1 96.1 91.2
LHR(highest) [12] 5 to 27 - - - - 100 91.2 97.4 100 100
  1. Current results show 4.2% improvement over our previous method using ICGA-ELM. All other multi-class and binary data sets are classified with 100% accuracy (shown in bold). Genes selected by BCGA-ELM (for all data sets) are classified using WEKA [19] machine learning package. These results are much lower for GCM multi-class data but are fairly consistent for other data sets compared to BCGA-ELM and other results in literature. (*σ is the variance).