سال انتشار: ۱۳۹۳
محل انتشار: کنفرانس بین المللی اقتصاد، حسابداری، مدیریت و علوم اجتماعی
تعداد صفحات: ۱۳
Mohammad Kadkhoda – Academic Staff, Mathematics & Informatics Research Group,ACECR at Tarbiat Modares University, P. O. Box: 14115-343 Tehran, Iran
Rozbeh Serri – Faculty of Electrical & Computer Engineering Science & Research Branch, Islamic Azad University Tehran, Iran
Stock exchange is one of the main financial sources which its main feature is continuous production of different and various financial data. Here we have presented a comprehensive datamining approach to predict stock index in Tehran exchange. Basically, stock index influences from many known and even unknown variables. Firstly, we have tried to extract the type and way of effective patterns by reviewing determining indices in main stock index fluctuations. Initially we have suggested the beneficial structure for available data. Then we used an approach based on MLP neural networks to extract the hidden knowledge. Suggested approach which is named multi-section multi-layer perceptron (MSMLP) has searched in the hidden space among data using some of MLP networks and trains to predict main index of stock and different groups’ stock index. After modeling the suggested approach, we have practically implemented it on the set of Tehran exchange data from 90.1.1 to 92.6.31, in a 30 months’ timescale. Then we have evaluated its capability to predict main index of Tehran exchange in the October 2013. Results obtained from implementation of MSMLP architecture indicated the 36.6 percent error to predict the stock index.