Rice cultivation is the major source of earning revenues worldwide. The productivity and yield of rice crops mainly depend on soil water balance and soil fertility. Irrigation water requirement (IWR) analysis helps to retain appropriate soil water balance and judiciously allocate water resources considering vegetative, reproductive, and ripening stages of rice growth. To restore fertility, the application of fertilizers is inevitable but most of these are squandered owing to improper fertilizer selection without evaluation of soil macro-nutrients. So, the enhancement of rice yield demands the well-balanced application of fertilizers along with specific IWR analysis in each growth stage. In this paper, eXtreme Gradient Boosting (XGBoost) is used to extract high-scoring, correlated environmental parameters with IWR. Stacking-based ensemble learning is used to predict evapotranspiration since it is a very crucial indicator of rice water demand in different growth stages. Based on selected features of XGBoost and predicted evapotranspiration, IWR specific to all the rice growth stages is predicted using the Bayesian genetic algorithm (\(Bay_{GA}\)) hyper-tuned random forest (RF). The parameters of the maximum and the minimum number of samples required to be at the leaf node of RF are hyper-tuned using \(Bay_{GA}\) to optimize performance. Comparative results indicate that IWR prediction using \(Bay_{GA}-RF\) outperforms other methods with Accuracy (86.12, 92.42, 91.24), MSE (0.182, 0.162, 0.196), RMSE (0.426, 0.402, 0.442), MAE (0.193, 0.174, 0.205) and NSE (0.911, 0.952, 0.944) in Vegetative, Reproductive and Ripening rice growth stages and accuracy of \(98\%\) to predict suitable fertilizer depending on Nitrogen, Phosphorous, and Potassium soil macro-nutrients.

水稻种植是全球收入的主要来源。水稻作物的生产力和产量主要取决于土壤水分平衡和土壤肥力。灌溉需水量 (IWR) 分析有助于保持适当的土壤水分平衡，并根据水稻生长的营养、生殖和成熟阶段明智地分配水资源。为了恢复肥力，化肥的施用是不可避免的，但由于没有对土壤常量养分进行评估而选择不当的肥料，大部分化肥被浪费了。因此，水稻产量的提高需要在每个生长阶段均衡施用肥料以及特定的 IWR 分析。在本文中，eXtreme Gradient Boosting (XGBoost) 用于提取高分、与 IWR 相关的环境参数。基于堆叠的集成学习用于预测蒸散量，因为它是不同生长阶段水稻需水量的一个非常重要的指标。基于 XGBoost 的选定特征和预测的蒸散量，使用贝叶斯遗传算法预测所有水稻生长阶段特有的 IWR（\(Bay_{GA}\) ) 超调随机森林 (RF)。使用\(Bay_{GA}\)对 RF 叶节点所需的最大和最小样本数参数进行超调，以优化性能。比较结果表明，使用\(Bay_{GA}-RF\)的 IWR 预测在精度（86.12、92.42、91.24）、MSE（0.182、0.162、0.196）、RMSE（0.426、0.402、0.442）、MAE（ 0.193, 0.174, 0.205) 和 NSE (0.911, 0.952, 0.944) 在营养、繁殖和成熟水稻生长阶段和\(98\%\)的准确度根据氮、磷和钾土壤大量养分预测合适的肥料.