<![CDATA[This research examines the application of Deep Learning (DL) technologies in mathematics and science education, focusing on technological approaches, pedagogical integration, and educational impacts. Following PRISMA 2020 guidelines, 225 studies published between January 2020 and February 2026 were analyzed from Scopus, Web of Science, PubMed, and SINTA databases. Results indicate that Convolutional Neural Networks (34.2%) and Recurrent Neural Networks/Long Short-Term Memory models (28.9%) dominate STEM applications, primarily implemented through Intelligent Tutoring Systems and adaptive learning platforms. Pedagogically, DL tools align predominantly with adaptive learning (38.7%) and inquiry-based approaches (34.2%). Evidence suggests positive impacts on learning outcomes (82.7% of studies reported significant improvements) and Higher-Order Thinking Skills, particularly critical thinking and problem-solving. However, implementation challenges persist, including technical infrastructure limitations (41.3%), data privacy concerns (36.9%), and insufficient teacher readiness (29.8%). This review concludes that while Deep Learning offers transformative potential for personalized STEM education, successful integration requires addressing ethical considerations, developing explainable AI systems, and enhancing educator preparation. Future research should prioritize longitudinal studies and equitable access to ensure DL technologies genuinely enhance rather than hinder mathematics and science learning experiences.]]>
Copyrights © 2026
