<![CDATA[This study investigates the effectiveness of deep learning based personalization in optimizing cognitive load and enhancing learning efficiency in primary science education. Drawing upon Cognitive Load Theory, the research addresses how advanced AI models specifically Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) architectures can dynamically adjust instructional content to meet learners’ cognitive needs in real time. Using a quasi-experimental design, 50 Grade 5 students were divided into AI and control groups, with the intervention delivered via an AI-enhanced Learning Management System (LMS). Quantitative findings reveal a 28.5% reduction in extraneous cognitive load and a 22.2% increase in germane cognitive load among the AI group, alongside higher post-test performance and a superior learning efficiency index (0.72 vs. 0.48). These outcomes suggest that the AI-driven system effectively minimized unnecessary processing while fostering deeper engagement and schema construction. Qualitative data from classroom observations and student interviews further support these results, highlighting increased learner autonomy, metacognitive awareness, and instructional responsiveness. Teachers benefited from real-time analytics, enabling more adaptive and differentiated instruction. The study concludes that deep learning personalization not only improves cognitive efficiency but also transforms the instructional landscape by supporting more equitable, individualized, and cognitively attuned science learning environments. These findings offer critical implications for digital pedagogy, curriculum design, and AI integration in STEAM education, particularly in underrepresented or early learning contexts. By reengineering science instruction through intelligent technologies, this research contributes to the development of future-ready, inclusive education systems grounded in cognitive science and data-informed personalization.]]>
