2692-2967ISSN (Online)
2770-0216ISSN (Print)
Achieving the United Nations Sustainable Development Goals (SDGs) requires robust forecasting tools capable of capturing complex temporal and multi-dimensional patterns in global sustainability data. Traditional statistical models often struggle with the high dimensionality and nonlinear dynamics of such datasets, motivating the adoption of advanced Deep Learning (DL) methods combined with metaheuristic optimization techniques. This paper proposes a novel forecasting framework leveraging Gated Recurrent Units (GRUs), Long Short-Term Memory networks (LSTMs), Recurrent Neural Networks (RNNs), and Convolutional Neural Networks (CNNs), optimized using the Human-Inspired Metaheuristic Optimization Algorithm (iHOW) and its binary variant (biHOW) for feature selection. The key contribution lies in integrating metaheuristic-driven feature selection and hyperparameter tuning to significantly enhance predictive performance and computational efficiency in SDG forecasting. Results highlight substantial improvements over baseline models: the GRU baseline achieved an R2 of 0.8037 with a Mean Squared Error (MSE) of 0.0772; application of biHOW for feature selection improved the GRU’s performance to an R2 of 0.9251 and MSE of 0.0011; and further hyperparameter tuning with iHOW elevated performance to an R2 of 0.9671 with MSE maintained at 0.0011. These results demonstrate the effectiveness of iHOW in balancing exploration and exploitation, providing high-accuracy forecasts with reduced error, thereby supporting more informed decision-making. The implications extend beyond sustainability analytics, presenting transferable forecasting frameworks for data-driven, real-time decision support in business sectors such as finance, energy, healthcare, and climate risk management. This alignment of predictive analytics with strategic financial and operational planning underscores the commercial value of integrating AI-driven forecasting into sustainability-focused investment and policy frameworks.
Read MoreDoi: https://doi.org/10.54216/AJBOR.130101
Vol. 13 Issue. 1 PP. 01-23, (2025)
In today’s interconnected global economy, accurate financial forecasting is critical for strengthening corporate decision-making, mitigating investment risks, and maintaining competitive advantage over the long term. Traditional forecasting models often struggle with the complexities of high-dimensional and nonlinear financial data. To address this challenge, we present a hybrid forecasting framework that integrates advanced machine learning techniques with an intelligent optimization algorithm. Specifically, the model combines Long Short- Term Memory (LSTM) networks with the Football Optimization Algorithm (FbOA) to optimize key features and tuning parameters. This approach yields more stable, efficient, and accurate financial predictions using a compact set of influential variables. The proposed framework offers a cost-effective solution for corporate finance applications, enhancing investor confidence and supporting strategic economic development. By bridging cutting-edge AI methodologies and practical financial analytics, this study highlights the transformative potential of hybrid models in reshaping financial forecasting in dynamic markets.
Read MoreDoi: https://doi.org/10.54216/AJBOR.130102
Vol. 13 Issue. 1 PP. 24-41, (2025)
Accurate exchange rate prediction is a critical challenge in financial forecasting, as fluctuations in exchange rates directly impact trade balances, investment strategies, and monetary policy decisions. Motivated by the need for robust and precise forecasting models, this study presents a novel framework that integrates deep learning (DL) methodologies with advanced metaheuristic optimization. At the core of this framework is the Continuous-Time Sequence Model (CTSM), complemented by the binary Ninja Optimization Algorithm (bNiOA) for feature selection and the Ninja Optimization Algorithm (NiOA) for hyperparameter tuning. Experimental results demonstrate substantial improvements in predictive performance. The baseline CTSM model achieved an accuracy of 0.8168 with a mean squared error (MSE) of 0.0718. After applying the bNiOA-driven feature selection, accuracy increased markedly to 0.9576, while the MSE was reduced to0.00067. Further optimization of hyperparameters through NiOA elevated the model’s accuracy to 0.9963, with an MSE of 0.00088. These results validate that the proposed optimization-enhanced deep learning pipeline effectively reduces feature redundancy and dimensionality, while finely tuning model parameters to achieve superior accuracy and generalization. The implications of this study are significant, providing policymakers, investors, and businesses with a powerful tool for risk management, strategic planning, and informed decision-making in volatile currency markets.
Read MoreDoi: https://doi.org/10.54216/AJBOR.130103
Vol. 13 Issue. 1 PP. 42-64, (2025)