K. Woods and K. W. Bowyer, "Generating ROC curves for artificial neural networks," IEEE Trans. Med. Imag., vol. 16, pp. 329--337, June 1997.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....relative preferences concerning the sensitivity or specificity of the detection task, one curve is typically preferred over the other. output threshold is usually varied to generate a ROC curve for ANN s [22] Traditionally, the classifier is trained prior to the generation of the ROC curve [22] [23]. In this situation, all but one point on the ROC curve represent operating points other than the one to which the classifier was naturally trained. A ROC curve that was generated with the same dataset that was used to train the classifier is referred to as a consistency ROC curve. A validation ....

....a conventional optimization that was trapped in a local minimum. The conventional ROC curves were generated by varying the output bias weight value, which corresponds to one component of . This is equivalent to varying the neural network output threshold. It should be noted that Woods and Bowyer [23] studied the effect of varying weight values other than the output bias weight in generating ROC curves. Their study concluded that varying a subset of the weights can produce better ROC curves than the ROC curves produced by varying the output threshold, as is conventionally done. By applying the ....

[Article contains additional citation context not shown here]

K. Woods and K. W. Bowyer, "Generating ROC curves for artificial neural networks," IEEE Trans. Med. Imag., vol. 16, pp. 329--337, June 1997.

....ANOVA does not account for case sample variance, one cannot generalize experimental results to the population from which the data were drawn. 1. Introduction Receiver Operating Characteristic (ROC) analysis [16] has proven invaluable for empirical studies of machinelearning algorithms (e.g. [14, 4, 18, 9]) Researchers have typically used analysis of variance, or ANOVA [15] to determine whether results from ROC analysis are statistically significant [4, 8] Yet, in the medical decision making community, which has a long tradition of conducting research on ROC analysis, the prevailing method in ....

K. Woods and K. Bowyer. Generating ROC curves for artificial neural networks. IEEE Transactions on Medical Imaging, 16(3):329--337, 1997. 4

....negatives) is a standard method for performance evaluation of detection algorithms. Typically, performance of the detection method is measured by the area under the ROC curve (A z ) ideally, A z = 1. Methods of generating the ROC curves for evaluation of ANN performance are described in [38] [39]. In addition to classification of ANN applications in breast cancer diagnosis based on the source of the input data, another classification can be done according to the type of ANN used. Majority of the methods employ multilayer perceptron architecture, usually threelayered structure, with ....

K. Woods and K.W. Bowyer, "Generating ROC Curves for Artificial Neural Networks," IEEE Trans. Medical Imaging, vol. 16, no. 3, pp. 329-337, June 1997.

....As a result, the majority of this chapter is devoted to this subject. The following sections explain how ROC curves are generated for each of the classifiers described in Chapter 4. Section 7.4 discusses our algorithm for generating ROC curves for ANNs. A comparative evaluation of our algorithm [94] has shown it to be superior to the currently accepted method of generating ROC points for backpropagation ANNs in terms of AUC, and the number and range of ROC points obtained. 7.1 Bayesian Classifiers Training the Bayesian classifiers (see Chapter 4) consists of computing the parameters (mean ....

....produced by the output node. Our method scales the weight on the bias inputs for selected nodes on the first hidden layer. The following two subsections describe the mechanics of our algorithm and provide some theoretical justification. A more complete set of experimental results can be found in [94]. 67 7.4.1 An Algorithm for ROC Curve Generation Consider a basic trained network. At each node, the weighted sum of inputs to that node becomes the input to a sigmoid function which determines the output value for the node: Output = 1 1 e W 1 X 1 W 2 X 2 : W d X d W 0 (7.2) In this ....

K. S. Woods and K. W. Bowyer, "Generating ROC curves for artificial neural networks," in Digital Mammography: Proceedings of the 2nd International Workshop on Digital Mammography, vol. 1069 of International Congress Series, (York, England), pp. 335--344, Elsevier Science B. V., July 10-12 1994.

No context found.

WOODS, K. Generating ROC curves for artificial neural networks. IEEE Transactions on medical imaging 16, 3 (June 1997), 329--337.

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