Variable Selection from Image Texture Feature for Automatic Classification of Concrete Surface Voids

Figure 6

Comparative experimental results on the selected 20 variables during one-time sample division. Red solid lines show the accuracies of a base classifier on 30% of testing samples in each round of resampling. Black dotted lines illustrate the corresponding mean accuracy values of the 1000 rounds of resampling. Blue dotted dash lines show the accuracies on 10% of independent testing samples. (a) The experimental results on the selected 20 variables. (b) The experimental results on 1000 rounds of random variable selection in 20 dimensions from the 21th variable to the 108th one. (c) The experimental results on 1000 rounds of random variable selection in 20 dimensions from all the 108 components. (d) The experimental results on all the 108 variables.

(a)

(b)

(c)

(d)