Understanding and managing soil productivity is crucial for sustainable agriculture and environmental protection. In this paper, we propose a new method, RL-RNN (Reinforcement Learning—Recurrent Neural Network), to analyze soil fertility levels. Soil fertility, which refers to the ability of the soil to promote plant growth, is influenced by a variety of factors including nutrient content, pH levels, and natural resources. Traditional methods of detecting soil crops usually require time-consuming and labor-intensive laboratory experiments; large scale may not be feasible for time tracking. Our proposed RL-RNN method is used to analyze geological data and accurately predict the rate of yield. The reinforcement learning model enables the model to learn from interactions with the environment and change its behavior based on the feedback. Coupled with recurrent neural networks that can process data sequentially, our method is able to capture time dependence in soil properties and make more accurate predictions. Combining RL and RNN, our observations are learned to improve soil fertility selection over time and continue to improve its performance empirically. These dynamic changes enable farmers and land managers to make informed decisions about land management practices, such as crop rotation for increased agricultural productivity and environmental sustainability. Through experiments and results, we demonstrate the effectiveness of RL-RNN in accurate soil fertility estimation and provide promising solutions. Our model has achieved a promising accuracy of 93.2%, recall score of 93.2%, precision score of 91.6%, and F1-score of 92.4%.