Climate change, resource scarcity, and population expansion pose serious difficulties to agriculture. By using sophisticated yield prediction and crop management techniques, machine learning (ML) presents viable answers to these problems. Through the use of genetic data, past crop yields, weather patterns, soil characteristics, and farming methods, this systematic literature review (SLR) investigates how ML might improve agricultural efficiency, resilience, and sustainability. Although both ML and deep learning (DL) techniques have been used extensively, DL models—which are distinguished by their neural network architectures—are becoming more and more popular because of their higher predictive accuracy. According to this review, temperature is the most commonly used variable that affects agricultural productivity estimates, followed by rainfall and soil quality. The main evaluation parameter for evaluating model performance turned out to be Root Mean Square Error (RMSE). However, given that many researchers emphasize limited access to public data, the scarcity of comprehensive and varied datasets continues to be a significant concern. The results highlight how crucial it is to create reliable datasets that cover a range of meteorological conditions, crop kinds, and yield records throughout time in order to improve machine learning-based agricultural production models. Addressing these problems would allow ML to reach its full potential in changing agriculture.