Table 2.

Accuracy of classifiers applied to prostate cancer data set excluding benign prostate hyperplasia samples.

	Classifier
	Unweighted	Weighted	Random forest	SVM
LOOCV
Top 1%	88.3a	93.3c	93.3h	95.0h
Top 5%	98.3b	96.7a	91.7i	95.0i
60:40 partitions
Top 1%	84.8 ± 10.2d	87.5 ± 8.6f	89.6 ± 7.3j	91.7 ± 6.8j
Top 5%	76.2 ± 11.4e	81.3 ± 10.3g	89.5 ± 7.3k	91.5 ± 6.4k

Notes: Accuracy of voting classifiers (unweighted and weighted), random forest and SVM applied to the prostate cancer data set excluding benign prostate hyperplasia samples from the control group. Features to include in the classifiers were derived using the top 1% or 5% of features based on t-statistics through a jackknife procedure using training sets in leave-one-out cross validation (LOOCV) or multiple random validation (60:40 partitions). Mean ± SD accuracy reported for 1,000 60:40 random partitions.

^aHighest accuracy achieved with 7 features in classifier;

^bHighest accuracy achieved with 9 features in classifier;

^cHighest accuracy achieved with 13 features in classifier;

^dHighest accuracy achieved with 21 features in classifier;

^eHighest accuracy achieved with 47 features in classifier;

^fHighest accuracy achieved with 23 features in classifier,

^gHighest accuracy achieved with 51 features in classifier. The number of features used in random forest and SVM varied across the training:test set partitions. The ranges were:

^h265–340 features;

ⁱ1,194–1,268 features;

^j212–533;

^k1,412–1,970 features.