NEURON OPTIMIZATION OF EVOLUTIONARYARTIFICIAL NEURAL NETWORKS FOR STOCK PRICEINDEX PREDICTION

This study presents an optimization procedurefor the number ofprocessing elements (neurons) of hidden layers to predicta stock priceindex using Evolutionary Artificial Neural Networks (EANN), inparticular, for the Istanbul Stock Market price index (ISE) in order tocontribute to the development of Intelligent Systems Methods formodeling several systems that are highly non-linear and uncertain.The US dollars/Turkish Lira (US/TRY) exchange rate, Euro/TurkishLira (EUR/TRY) exchange rate, ISE National 100 (XU100) index,world oil price, and gold price were used as for a period ofapproximately 10 years’ daily data asinputs. Performance isbenchmarked by mean squared error, normalized mean squarederror;mean absolute error and thecorrelationcoefficient.Withthe fixedneural network architecture and optimized parameters, evolutionaryneural networks perform better performance valueswhen thenumberof neurons used in hidden layers isoptimized