Selection of Attributes

A value between 0 and 1 was obtained for each attribute, after performing attribute weighting models on the Mds. This value shows the relevance of the attribute with regards to the imprinted or biallelic expressed gene as a target attribute. Variables were selected with weights more than 0.50 and consequently 11 new datasets were created (Awds) (Table 1). These data sets were named based on their attribute weighting models (Info Gain, Info Gain Ratio, PCA, Correlation, Rule, Deviation, chi squared, Gini index uncertainty, relief and SVM) and then supervised and unsupervised models were used. Each model of the supervised or unsupervised algorithm was performed 12 times, first on the Mds and then on the new eleven datasets (Awds).

Seqlength = length of sequence, CpGi = CpG island, CpGi gene/kb = CpG island per kb of gene region, CpGn = number of CpG dinucleotide, CpGn gene/kb = number of CpG dinucleotide per kb of gene region, CpGn exons = number of CpG dinucleotide in exon region, LINE/kb Gene = number of LINE elements per kb gene region, SINE/kb Gene = number of SINE elements per kb gene region, LINE+SINE+LTR /kb Gene = number of LINE and SINE and LTR per kb gene region, 1–10–100 kb UP and Dwn = 1, 10, 100 kb up and down stream, simplrepExones = simple repeat in exon region.