An Interpretable AI-Based Optimization Paradigm for Scalable and Low-Power Intelligent Systems

Abstract

The rapid growth of intelligent systems and edge computing applications has created a significant demand for scalable and energy-efficient optimization frameworks capable of delivering high predictive performance while maintaining interpretability. Conventional deep learning approaches often suffer from high computational complexity and limited transparency, making them unsuitable for low-power intelligent environments. To address these challenges, this paper proposes an interpretable AI-based optimization paradigm for scalable and low-power intelligent systems, integrating explainable machine learning mechanisms with adaptive optimization strategies to enhance decision-making capability and computational efficiency. The proposed framework employs feature extraction, interpretable learning modules, and adaptive optimization techniques to minimize energy consumption while maximizing prediction accuracy. Experiments were conducted on benchmark datasets consisting of 25,000 samples, and the performance was evaluated using accuracy, precision, recall, F1-score, computational latency, and power consumption metrics. The proposed model achieved an accuracy of 98.74%, precision of 98.41%, recall of 98.62%, and F1-score of 98.51%. Furthermore, the framework reduced computational latency to 12.8 ms, decreased memory utilization by 31.6%, and lowered power consumption by 38.4% compared with conventional deep learning approaches. Scalability analysis demonstrated stable performance with an average throughput improvement of 27.9% under increasing workloads. The obtained results confirm that the proposed interpretable AI-based optimization paradigm provides an effective balance between accuracy, transparency, scalability, and energy efficiency. Therefore, the framework represents a promising solution for next-generation low-power intelligent systems, including edge devices, Internet of Things (IoT) platforms, and resource-constrained artificial intelligence applications.