Agriculture is essential for food production and raw materials. A key aspect of this sector is harvest, the stage at which the commercial part of the plant is separated. Timely harvesting minimizes post-harvest losses, preserves product quality, and optimizes production processes. Globally, a substantial amount of food is wasted, impacting food security and natural resources. To address this problem, an Adaptive Neuro-Fuzzy Inference System was developed to predict timely harvesting in crops. Stevia, a native plant from Brazil and Paraguay, with an annual production of 100,000 to 200,000 tons and a market of 400 million dollars, is the focus of this study. The system considers soil pH, Brix Degrees, and leaf colorimetry as inputs. The output is binary: 1 indicates timely harvest and 0 indicates no timely harvest. To assess its performance, Leave-One-Out Cross-Validation was used, obtaining an $r^2$ of 0.99965 and an Absolute Residual Error of 0.00064305, demonstrating its accuracy and robustness. In addition, an interactive application that allows farmers to evaluate crop status and optimize decision-making was developed.

Agriculture is essential for food production and raw materials. A key aspect of this sector is harvest, the stage at which the commercial part of the plant is separated. Timely harvesting minimizes post-harvest losses, preserves product quality and optimizes production processes. Globally, a substantial amount of food is wasted, impacting food security and natural resources. To address this problem, an Adaptive Neuro-Fuzzy Inference System was developed to predict timely harvesting in crops. Stevia, a native plant from Brazil and Paraguay, with an annual production 100,000 to 200,000 tons and a market of 400 million dollars, is the focus of this study. The system considers soil pH, Brix Degrees and leaf colorimetry as inputs. The output is binary: 1 indicates timely harvest and 0 indicates no timely harvest. To assess its performance, Leave One Out Cross Validation was used, obtaining an r² of 0.99965 and a Residual Absolute Error of 0.00064305, demonstrating its accuracy and robustness. In addition, an interactive application that allows farmers to evaluate crop status and optimize decision making was developed.