Augmented Reality (AR) has become a game changer in the preservation and appreciation of cultural heritage, increasing user engagement through immersive experiences. With the rise of mobile computing, integrating Machine Learning (ML) into AR applications enabled real-time analysis, object recognition, and adaptive content delivery. This systematic review explores ML frameworks like TensorFlow Lite, PyTorch Mobile, and Caffe for augmented reality (AR) applications in cultural heritage, with an emphasis on performance metrics, resources, and implementation considerations. With the idea of systematically analyzing and comparing the performance metrics of TensorFlow Lite, PyTorch Mobile, and Caffe in AR applications, and evaluating the technical requirements of its implementation as a recommendation basis for the framework selection mentioned. The primary challenges, such as hardware constraints, data quality, and user experience optimization, are also discussed. This review offers researchers and developers the opportunity to improve mobile AR experiences by highlighting trends and gaps in the field, with the goal of implementing MLdriven solutions for cultural heritage. © 2025 IEEE.