Asian Journal of Research in Computer Science

The rapid advancement of artificial intelligence (AI) has transformed multiple sectors; however, the growing computational complexity of modern AI models has raised serious concerns regarding energy consumption and carbon emissions. Addressing this challenge, this paper proposes a sustainable AI framework that integrates energy-efficient algorithm design, resource-aware model optimization, and environmentally conscious hardware utilization. Sustainability is treated as a core design principle rather than a secondary optimization goal.

To demonstrate the practical applicability of the proposed framework, a case study on Air Quality Index (AQI) prediction is presented using a lightweight Multi-Layer Perceptron (MLP) model implemented in PyTorch. The model adopts a shallow feedforward architecture with ReLU activation, dropout regularization, and normalized inputs, and is trained entirely on a standard CPU to minimize energy usage. Mean Squared Error (MSE) was employed as the evaluation metric, with the proposed model achieving a test MSE of 129.31, indicating stable predictive performance despite limited computational resources. The results highlight that competitive accuracy can be achieved without relying on energy-intensive GPUs or complex deep learning architectures.

By aligning AI development with environmental responsibility, this work demonstrates that sustainable model design can effectively support climate-aware applications such as air quality monitoring. The proposed framework contributes toward achieving broader sustainability objectives, including reduced carbon footprint and alignment with the United Nations Sustainable Development Goals. Overall, the study emphasizes that environmentally responsible AI systems are both feasible and essential for long-term technological and ecological sustainability.