![]() | Only 14 pages are availabe for public view |
Abstract A detailed review of 48 research papers proposing market price prediction methodologies focused on artificial neural networks is given in this thesis. Here, the reported research is categorized on the basis of various prediction techniques. In addition, the studies are analysed on the basis of the datasets used, metrics of performance assessment and targets for prediction. Collective data shows that the forecast of the stock price involves multiple considerations that need to be approached effectively and specifically which. All this prompted us to apply different machine learning methods in this field, taking into account all these considerations as shown in this thesis. This thesis investigates usefulness of three Deep Learning (DL) strategies for training nonlinear autoregressive network with exogenous inputs (NARX), including Bayesian Regularisation (BR), Levenberg-Marquardt (LM), and Scaled Conjugate Gradient (SCG) to precisely predict the closing price of the Egyptian Stock Exchange (EGX-30, EGX-30-Capped, EGX-50-EWI, EGX-70, EGX-100, and NIlE) indices. An analytical comparison is developed between the prediction models experimented, taking into account all techniques for the 1-day, 3-day, 5-day, 7-day, 15-day and 30-day time horizon in advance, applicable to all the datasets used in this analysis. Statistical methods such as Mean Squared Error (MSE) and R correlation are used for performance assessment. from the simulation outcome, it can be clearly indicated that BR outperforms other short-term predictive models , particularly for the next 3 days. On the other hand, for long-term prediction, LM produces better prediction accuracy than BR and SCG-based models , especially for 7-day prediction. Subsequently, a hybridised support vector regression (SVR) approach with Equilibrium Optimizer (EO) for Egyptian trade forecasting is proposed. Three indexes, such as EGX 30, EGX 30 capped and EGX 50 EWI, are used and modelled using technical analysis methods to estimate market closing prices. The effectiveness of the use of technical metrics and comparative measurements for forecasting purposes has also been measured. The suggested EO-SVR-based prediction model is observed and the performance of the proposed model was evaluated using Mean Absolute Percentage Error (MAPE), Average, Standard Deviation, Best Fit, Worst Fit and CPU time. A comparison with recently developed algorithms such as Whale Optimization Algorithm (WOA), Salp Swarm Algorithm (SSA), Harris Hawks Optimization (HHO), Gray Wolf Optimizer (GWA), Henry Gas Solubility Optimization (HGSO), Barnacles Mating Optimizer (BMO), Manta Ray Foraging Optimization (MRFO) and Slime Mould Algorithm (SMA) had been done. EO-SVR is an optimal model with superior performance, and we have found that there is no need for technical indicators and statistical measures to be used, as their influence is not significant. |