Citation Context

... most regression data are inherently inaccurate, e.g. due to experimental and measurement errors. REC curves were initially motivated by the ɛinsensitive loss function used in SVM regression methods (=-=Vapnik, 1995-=-). The ɛ-insensitive loss (max{0, |y− f(x)| − ɛ}) has proven to be very robust, but there is no a priori way to pick ɛ. The REC curve based on absolute error metric considers an ɛ-insensitive loss for...

Citation Context

...ints are sorted in ascending order. Then ɛm is the maximum observed error, and the frequency ˆ P (ɛi) = i/m. By the Glivenko-Cantelli lemma, ˆ P (ɛ) converges to P (ɛ) uniformly over all values of ɛ (=-=DeGroot, 1986-=-). We give a rough sketch of the argument about AOC assuming the full REC curve can be plotted in the figure, in other words, ɛm ≤ ˆɛ. The expectation of ɛ is defined as E(ɛ) = � ɛp(ɛ)dɛ. For any maxi...

Citation Context

...ction for classification problems (Egan, 1975; Swets et al., 2000; Fawcett, 2003). ROC curves address many of the limitations of comparing algorithms based on a single misclassification cost measure (=-=Provost et al., 1998-=-). An ROC curve characterizes the performance of a binary classification model across all possible trade-offs between the false negative and false positive classification rates. An ROC graph allows th...

Citation Context

...used to guide modeling. 1. Introduction Receiver Operating Characteristic (ROC) curves have proven to be a valuable way to evaluate the quality of a discriminant function for classification problems (=-=Egan, 1975-=-; Swets et al., 2000; Fawcett, 2003). ROC curves address many of the limitations of comparing algorithms based on a single misclassification cost measure (Provost et al., 1998). An ROC curve character...

Citation Context

...duction Receiver Operating Characteristic (ROC) curves have proven to be a valuable way to evaluate the quality of a discriminant function for classification problems (Egan, 1975; Swets et al., 2000; =-=Fawcett, 2003-=-). ROC curves address many of the limitations of comparing algorithms based on a single misclassification cost measure (Provost et al., 1998). An ROC curve characterizes the performance of a binary cl...

Citation Context

.... The models were constructed using classic regression SVM trained with different values of ɛ (Vapnik, 1995), linear partial least squares (PLS) (Wold, 1966), and kernel partial least squares (KPLS) (=-=Rosipal & Trejo, 2001-=-) 1 . All methods produced reasonable results that outperformed the null model. The results show that the SVM1 model performs better than the rest. The KPLS and the SVM2 models perform very similarly....

Citation Context

...e modeling. 1. Introduction Receiver Operating Characteristic (ROC) curves have proven to be a valuable way to evaluate the quality of a discriminant function for classification problems (Egan, 1975; =-=Swets et al., 2000-=-; Fawcett, 2003). ROC curves address many of the limitations of comparing algorithms based on a single misclassification cost measure (Provost et al., 1998). An ROC curve characterizes the performance...