Table 3.

Three subsets of feature methods of the training group predicting TNBC

Feature selection method	Parameter
Fisher	WavEnHH_s-3, WavEnLH_s-4, WavEnHL_s-4, WavEnHL_s-2, WavEnLH_s-3, GrMean, S(0,1) SumAverg, S(1,1) SumAverg, S(1,-1) SumAverg, S(2,0) SumAverg
POE + ACC	WavEnLH_s-4, Kurtosis, Perc.01%, Vertl_LngREmph, WavEnHH_s-5, Teta4, WavEnHL_s-5, 135dr_ShrtREmp, GrKurtosis, WavEnHH_s-1
MI	WavEnLL_s-2, WavEnLL_s-1, 135dr_Fraction, 135dr_LngREmph, WavEnLH_s-4, S(0,2) SumOfSqs, S(1,0) SumOfSqs, S(1,1) SumOfSqs, S(2,0) SumOfSqs, S(2,2) SumOfSqs

Fisher: Fisher parameter method; POE + ACC: Classification error rate combined average correlation coefficient method; MI: Related Information Measurement; Wavelet transform: WavEnHH_s-3,WavEnLH_s-4,WavEnHL_s-4, WavEnHL_s-2, WavEnLH_s-3, WavEnHH_s-5, WavEnHL_s-5, WavEnHH_s-1, WavEnLL_s-2, WavEnLL_s-1; Gradient model: GrMean, 135dr_ShrtREmp, GrKurtosis, 135dr_Fraction, 135dr_LngREmph; Gray Level Co-occurrence Matrix: S(0,1) SumAverg, S(1,1) SumAverg, S(1,-1) SumAverg, S(2,0) SumAverg, S(0,2) SumOfSqs, S(1,0) SumOfSqs, S(1,1) SumOfSqs, S(2,0) SumOfSqs, S(2,2) SumOfSqs; Histogram: Kurtosis, Perc.01%; Run matrix: Vertl_LngREmph; Autoregressive model: Teta4