This study presents an efficient approach to predict the Rabi and Kharif crop yield using a relatively new and robust machine learning (ML) model named random forest (RF). The standard precipitation evaporation index (SPEI) with different time lags (e.g., 1 to 12 months) are utilized as predictive variables. The SPEI was estimated using the climate prediction center (CPC) precipitation, and temperature dataset for the period 1981–2015 are employed. The feasibility of the RF model is validated against some other well-known ML models such as support vector regression (SVR), k-nearest neighbors (K-NN), and bagged CART models. The results showed a significant relationship between crop yields and the SPEI. The RF model showed the highest performance with the minimum values of absolute error measures (e.g., root mean square error (RMSE) and mean absolute error (MAE)) in the testing phase (0.1826–0.1383) and (0.1234–0.1092) for cotton and wheat production, respectively. Cotton yield prediction accuracy using the RF model improved compared to the SVR, K-NN, bagged CART, and ANN in terms of RMSE, and MAE indices are 12–10.79%, 12.33–10.79%, and 5.7–0.17%, respectively. Overall, the RF model provided a reliable alternative ML-based strategy for the cotton and wheat yield prediction over the Pakistan region.

本研究提出了一种使用名为随机森林 (RF) 的相对较新且稳健的机器学习 (ML) 模型来预测 Rabi 和 Kharif 作物产量的有效方法。具有不同时间滞后（例如，1 至 12 个月）的标准降水蒸发指数 (SPEI) 被用作预测变量。SPEI 使用气候预测中心 (CPC) 降水估算，并使用 1981-2015 年期间的温度数据集。RF 模型的可行性已针对其他一些众所周知的 ML 模型进行了验证，例如支持向量回归 (SVR)、k-最近邻 (K-NN) 和袋装 CART 模型。结果表明，作物产量与 SPEI 之间存在显着关系。RF 模型在绝对误差测量值（例如，棉花和小麦生产分别在测试阶段 (0.1826–0.1383) 和 (0.1234–0.1092) 的均方根误差 (RMSE) 和平均绝对误差 (MAE)。与 SVR、K-NN、袋装 CART 和 ANN 相比，使用 RF 模型的棉花产量预测精度在 RMSE 方面有所提高，MAE 指数分别为 12-10.79%、12.33-10.79% 和 5.7-0.17%。总体而言，RF 模型为巴基斯坦地区的棉花和小麦产量预测提供了一种可靠的基于 ML 的替代策略。