Did a t-test with Bonferroni correction on column "33" in class zero which wound up with a corrected p-value < 3 x 10^-8. Tests for normality looked spotless. No way this is sampling variability.
Actually, doing t-tests for differences in means for column "33" and column "65" between the two classes yield Bonferroni corrected p-values < 4 x 10^-7 and < 7 x 10^-4, respectively.
Did a t-test with Bonferroni correction on column "33" in class zero which wound up with a corrected p-value < 3 x 10^-8. Tests for normality looked spotless. No way this is sampling variability.