Predicting stock index using neural network combined with evolutionary computation methods

Price index forecasting is one of the most important problems in financial markets. In the past decades the prediction of stock index has played a vital role in the financial situation of several companies which have stocks in the market. In this paper we use Multi Layer Perceptron (MLP) neural network in stock index prediction. Three searching algorithms were used to get the best network architecture and parameters, increase the accuracy of prediction and decrease the training time. SA (simulated annealing), GA (genetic algorithm) and hybrid approach combining both SA and GA together are conducted and their results are compared in this study. The results showed that the best algorithm in terms of accuracy is SA which was found to be more accurate than GA by 40% and more accurate than hybrid approach by 30%. Concerning training time the best algorithm was found to be SA which takes in average 7 minutes in training. On the contrary GA takes in average 73 minutes and finally the hybrid approach takes in average 98 minutes. All the tests were carried on a training set containing 2000 records.