Prediction of Stock Market Prices Using Hybrid Intelligent System رسالة ماجستير

Abstract

The main aim of the stock market is to provide a safe environment for trading in order to achieve better services to investors and maintain their investments. The stock market in any country plays an important role in the national economic development and gross domestic product (GDP). Hence, the process of the prediction of the stock market returns is of high importance attention. If the stock market predictions come true, they will definitely help the investors formulate their investment strategies on solid and scientific grounds. Consequently, the concepts of the stock market parameters that affect the market that we have to consider with the mathematical methods used to predict the econometrics emerge and to quantify the uncertainty through multivariate causal criteria with the support of the randomization of the market. The variables of the market like share price, interest rates, exchange rates, inflation, etc, can be predicted using intelligent and smart methods, which allow the investors to invest their money in a more secure and reliable way. In this thesis, we presented an optimized hybrid model that combines the multilayer perceptron neural networks with genetic algorithms (MLPNNs-GAs) to predict the status of the Palestinian stock market on Al-Quds Index as the main indicator. In addition, the stock market data of three biggest Palestinian companies (Paltel, Padico, and Bank of Palestine) will be used to predict its stock prices. In this research, the idea behind combining artificial neural networks (ANNs) with GAs is that characteristics of data in stock prices have high volatility, nonlinear in type. And as we know, without imposing a particular relationship in the data, ANNs has the ability to learn unobserved relationships in the data. Genetic algorithms (GAs) are used to optimize the weights for the NNs, and GAs will pick the best weights in order to optimize performance and get the best-predicted minimum mean square error (MSE) value. The GAs process applied using the best combination methods of the GAs main steps. Furthermore, we applied another two models of different neural networks VI methodologies; multilayer perceptron neural networks trained using Levenberg-Marquardt back propagation (MLPNNs-LM) and the recurrent neural networks RNNs-LM trained using Levenberg-Marquardt back propagation too and compared the performance of the three applied models in term of MSE. The experimental results obtained from the proposed MLPNNs-GAs model and the other applied NNs models (MLPNNs-LM and RNNs-LM), showed that the performance of the hybrid model (MLPNNs-GAs) outperforms the MLPNNs-LM and RNNs-LM models in forecasting the closed price of the 4 datasets which present the stock market (Padico, Paltel, Palestine Bank, and the Al-Quds index), where the MLPNNs-LM model produce better accuracy than RNNs-LM in general.