Training a single multiplicative neuron with a harmony search algorithm for prediction of S&P500 index - An extensive performance evaluation

Harmony Search is a relatively new meta-heuristic algorithm for continuous optimization, in which its concept imitates the process of music improvisation. This paper applied an improved harmony search algorithm called Harmony Search with Adaptive Pitch Adjustment (HSAPA) for prediction of stock market index. HSAPA is applied to optimize the weights and biases of Single Multiplicative Neuron for the prediction of daily S&P500 index. Its prediction performance has been extensively evaluated using various sizes of dataset, training proportions, and beginning dates spanning from 1990 to 2009, a totaling of 108 test sets. The prediction results are compared to those of standard Back Propagation learning method and Opposition-based Differential Evolution algorithm, a very efficient and widely-accepted evolutionary algorithm. The results demonstrate that HSAPA is very promising for the stock market index prediction, measured with the mean absolute percentage error of the prediction results.