Deep Hybrid Learning for Sustainable Industrial Forecasting: Integrating CNN–LSTM Models to Enhance Economic Efficiency and Carbon Performance

This paper explores the contribution of neural network-based safeguarding models to enhancing the environmental resilience and economic efficiency of industrial supply chains. The methodology includes a review of existing literature for a quasi-experimental study conducted from the perspective of a manufacturer. Using this approach, the study analyzes the transition from traditional statistical safeguarding practices to modern neural predictive frameworks, the amount of data available, and assesses their impact on decision-making and overall chain performance. The results from a Tunisian organization indicate that deep hybrid training architectures, particularly CNN-LSTM models, significantly improve the accuracy of demand forecasting, resulting in concurrent gains in operational efficiency and environmental performance. The organization also achieved a reduction in its annual costs of 2.25 million Tunisian dinars, leading to a decrease in carbon emissions. The study also identifies key obstacles, such as the fragmentation of data infrastructure, the lack of digital skills, and global development costs, which necessitate the effective adoption of deep training. Based on these findings, the paper proposes a dual-performance neural network framework to help managers and policymakers align technological innovation with the realities of emerging economies.