DEVELOPMENT OF ENERGY EFFICIENT ARTIFICIAL INTELLIGENCE MODELS FOR EDGE COMPUTING BASED REAL TIME APPLICATIONS IN LOW POWER AND RESOURCE LIMITED IOT ENVIRONMENT

Abstract

The rapid expansion of Internet of Things (IoT) applications has created a growing demand for real-time intelligent processing; however, conventional cloud-based artificial intelligence (AI) systems face challenges related to high latency, excessive energy consumption, bandwidth limitations, and dependency on continuous connectivity. This research addresses the problem of developing energy-efficient AI models capable of operating effectively in low-power and resource-constrained IoT environments through edge computing-based architectures. The study aims to design and evaluate optimized AI models that balance computational performance, energy efficiency, and real-time decision-making capabilities for edge-enabled IoT applications. A systematic methodology was adopted involving lightweight deep learning architectures, model compression techniques, quantization approaches, and performance evaluation using energy consumption, processing latency, memory utilization, and prediction accuracy as key metrics. Experimental analysis demonstrates that optimized edge AI models significantly reduce computational overhead and energy requirements while maintaining competitive accuracy levels compared with traditional AI models deployed on centralized cloud platforms. The findings reveal that techniques such as model pruning, efficient neural network structures, and adaptive resource management enhance the sustainability and scalability of IoT-based intelligent systems. The research highlights the importance of integrating energy-aware AI algorithms with edge computing frameworks to support real-time applications in smart cities, healthcare monitoring, industrial automation, and environmental sensing. The proposed approach provides a practical pathway for developing sustainable, autonomous, and efficient AI-driven IoT ecosystems, particularly in environments where computational resources and energy availability are limited