Discovering biomarkers from gene expression data for predicting cancer subgroups using neural networks and relational fuzzy clustering

Table 5 Training and test error statistics with all possible subsets of six biomarkers

The maximum, minimum and the average training and test errors obtained using MLP networks with all possible subsets of size six genes (i.e., after removing one gene at a time from the set of identified marker genes). With each combination of 6 genes, the network was trained 20 times with different initializations and the statistics are computed based on the results of these 20 trials

ISSN: 1471-2105