An efficient system for stock market prediction

Abstract

This paper presents an efficient system for accurate, confident, general and responsive stock market prediction, employing Artificial Neural Networks (ANN). For technical indicators, Multi-Layer Perceptron (MLP) ANN is used and trained with Kullback Leibler Divergence (KLD) learning algorithm because it converges fast in addition to offering generalization in the learning mechanism. On the other hand, Radial Basis Function Neural Network (RBFNN) trained with Localized Generalization Error (L-GEM) is used for candlesticks patterns. The accuracy, generalization and statistical-significance of the developed system were confirmed through various local and international data sets. Next, sensitivity analysis was conducted for the different parameters that influence the system efficiency metrics. In order to have responsive prediction, the proposed system was evolved, employing concurrent programming to get benefit from the off-the-shelf multi-core architectures. Then, the performance of the developed system was evaluated to confirm acceptance scalability and utilization.