Financial and stock market forecasting is of significant importance for investors, stakeholders, researchers, and governments, as it aids in directing liquidity toward optimal channels and allocating scarce financial resources. Given the inherent volatility and unpredictability of the stock market, this high-risk environment, influenced by rapid price fluctuations driven by economic, political, and social factors, can lead to gains or losses for investors. Consequently, the ability to forecast market movements assists in reducing investment risk and making strategic decisions. In this study, an innovative approach based on directional learning and distance-error minimization is introduced for financial market forecasting. This approach not only reduces the quantitative discrepancy between actual and predicted values but also considers the directional movement of prices and the financial value of trades. In the proposed model, the objective function is adjusted by assigning weights to higher-value trades, aiming to improve prediction accuracy in profitable directions. The main goal of this thesis is to enhance the accuracy of predictions and the efficiency of financial decisions by applying a novel approach in finance. A multilayer perceptron model has been utilized to implement this algorithm, as it is adaptable to complex financial data. After modeling, the performance of the proposed model was eva‎luated using real data, with a detailed analysis conducted on its application and effectiveness. This method can improve prediction accuracy and reduce uncertainties and risks in financial markets. The application of the proposed neural network in analyzing financial datasets shows that the rate of return on investment based on this approach significantly surpasses that of two other models, namely, multivariate regression and the classic multilayer perceptron neural network. On average, the rate of return on investment with the proposed model is 1.88%, compared to 1.69% for the multivariate regression model and 1.77% for the classic multilayer perceptron neural network. These results indicate the superior performance of the proposed neural network over conventional methods. Consequently, the application of this model can be considered an effective solution in the field of financial analysis and investment.

مهدي خاشعي آشياني

صبا صارمي نيا , حسين خسروشاهي