Schwarzer G, Vach W, On the misuses of artificial neural networks for prognostic and diagnostic classification in oncology. Stat Med. 2000 Feb 29;19(4):541-61.  Home/Search   Document Details and Download   Summary   Related Articles

....ratios. There is also a growing concern in the data mining literature that published models are often over fitted (e.g. by choosing classifier parameters manually in a non blinded way to the test data, or by examining in a blinded fashion too many models using automated search procedures) [7,15]. To what extend are the reported results in microarray analysis an artifact of the evaluation methodologies employed What methods should be employed to rule out the possibility of poor methodology inflating estimates of performance And what factors explain these performance levels that, ....

.... test on cases that were not used for training to first perform gene selection, then optimize parameters for a classifier, and choose the best classifier [8] Thus we do not examine the effect of omitting cross validation since it is well established in the data mining and statistical literature [7,8,15] that such an omission leads to overestimated performance, and in the bioinformatics literature some form of regularization is always used. Similarly we do not investigate the effect of estimating classification performance using accuracy (a.k.a, 0 1 loss , a measure that is sensitive to the ....

Schwarzer G, Vach W, On the misuses of artificial neural networks for prognostic and diagnostic classification in oncology. Stat Med. 2000 Feb 29;19(4):541-61.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC