Walczak, S., and Cerpa, N. Heuristic principles for the design of artificial neural networks. Information and Software Technology, 41, 2 (1999), 107--117.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....that problems of configuring neural nets tend to rather counteract their superior performance in terms of accuracy. Clearly there is a need for further investigation, particularly in finding appropriate configuration heuristics for neural nets. Whilst some heuristics have been published (e.g. Walczak, and Cerpa, 1999), we unfortunately did not find them to be of great value for this particular prediction task. 19 Nevertheless, we believe that these ML methods warrant further investigation, particularly to explore under which conditions they are most likely to be effective. Acknowledgements This ....

Walczak, S. and N. Cerpa, (1999) 'Heuristic principles for the design of artificial neural networks', Information & Softw. Technol., 41(2), pp107-117, 1999.

.... two critical design issues still face financial traders desiring to use neural networks: selection of appropriate variables and capturing a sufficient quantity of training examples to permit the neural network to adequately model the financial time series [38, 39] Research by Walczak and Cerpa [32] and others [27, 39] has already focused on the question of selecting appropriate input values for modeling time series domains. However, the question of how much historical information is required to produce the best performing model has not been addressed by neural network researchers. The ....

....206 DATA REQUIREMENTS FOR FINANCIAL FORECASTING WITH NEURAL NETWORKS 207 mogeneous input variables. A preliminary effort to maximize the output performance of the developed backpropagation neural networks is conducted by ensuring adequate domain knowledge representation from the input variables [25, 27, 32] (since analyzing the effect of training set size on neural network models that did not provide any advancement over a standard random walk model would not be very interesting) Hence, following Walczak et al. s [34] suggestion that multiple time lags provide a significant trading advantage for ....

[Article contains additional citation context not shown here]

Walczak, S., and Cerpa, N. Heuristic principles for the design of artificial neural networks. Information and Software Technology, 41, 2 (1999), 107--117.

.... A reason why artificial neural network applications in medical domains have lagged behind non medical domains is the difficulty in determining which clinical variables are relevant [23] Identifying the variables that have a definite impact on a clinical decision or outcome is a critical task [32] and artificial neural networks cannot accurately model a domain problem without inclusion of all the relevant variable values. Another problem in medical domains is the scarcity of data. Classifier learning in general, and artificial neural network learning specifically, improves as ....

.... estimates the true error rate of the artificial neural network classification model [33] The final problem to be solved for artificial neural network solutions of medical domain problems is the selection of the artificial neural network learning method and subsequent hidden node architecture [32]. Backpropagation is by far the most widely recognized artificial neural network training algorithm [35,36] but recently radial basis function (RBF) artificial neural network solutions to medical domain problems have been reported. Comparisons of backpropagation and RBF neural networks in ....

[Article contains additional citation context not shown here]

S. Walczak and N. Cerpa, "Heuristic Principles for the Design of Artificial Neural Networks," Information and Software Technology, vol. 41, no. 2, pp. 109-119, 1999.

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