Table 3.

Performance of classifiers applied to independent validation set of lung cancer data set.

	Unweighted	Weighted	Random forest	SVM
Accuracy
LOOCV
Top 1%	99.3a	100b	98.7e	100e
Top 5%	99.3a	100b	98.7f	100f
60:40 partitions
Top 1%	98.7c	100d	99.3g	94.6g
Top 5%	98.7c	100d	100h	84.6h
Sensitivity
LOOCV
Top 1%	100	100	99.2	100
Top 5%	100	100	99.2	100
60:40 partitions
Top 1%	99.2	100	100	94.0
Top 5%	99.2	100	100	82.8
Positive predictive value
LOOCV
Top 1%	99.3	100	99.2	100
Top 5%	99.3	100	99.2	100
60:40 partitions
Top 1%	99.2	100	99.3	100
Top 5%	99.2	100	100	100

Notes: Accuracy, sensitivity, and positive predictive value of voting classifiers (unweighted and weighted), random forest and SVM applied to independent data sets from the lung cancer data set. 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).

^aHighest accuracy achieved with 37 features in classifier;

^bHighest accuracy achieved with 23 features in classifier;

^cHighest accuracy achieved with 15 features in classifier;

^dHighest accuracy achieved with 49 features in classifier. The number of features used in developing SVM and random forest classifiers were:

^e452 features;

^f1,791 features;

^g4,172 features;

^h9,628 features.