Today’s agronomy is data-rich, and machine learning (ML) provides the ability to efficiently predict crop yields, utilizing high-volume data to optimize agricultural decision-making. Numerous ML models are employed in yield prediction research, yet systemized know-how on its crop-targeted utilizations is lacking, specifically for soybean and maize, world’s vital crops. Henceforth, this systematic literature review (SLR) is performed to retrieve and consolidate the ML techniques and key features utilized in maize and soybean yield prediction research. Study’s search criteria utilized four electronic databases including ProQuest, Wiley, Science Direct, and EBSCOhost, totally producing 1859 related articles, which were finally reduced to 82 articles following SLR’s inclusion and exclusion criteria. Aligning with the study objectives, all papers were thoroughly analysed for generating common consensus and future research recommendations. The SLR analysis noted that ML is gaining popularity in the studied domain with a significant increase in its adoption after 2019. This study revealed the temperature, precipitation, historical crop yield, normalized difference vegetation index (NDVI), and soil pH to be the most utilized variables in ML for yield prediction research. The Random Forest (RF), Artificial Neural Networks (ANN), Support Vector Machines (SVM), and Extreme Gradient Boosting (XG-Boost) were identified as the mostly used ML algorithms. Most often applied deep learning (DL) techniques include long short-term memory (LSTM) and convolutional neural networks (CNN). In the utilized models, the most used performance assessment measures were noted as the coefficient of determination (R²), root absolute error (RAE), root mean square error (RMSE), and mean absolute error (MAE). Most applied software for building ML models includes Python, MATLAB, Weka, R, and SPSS. Altogether, there is a rising trend among ML researchers towards leveraging ensemble techniques in for the betterment of model performance and reliability.

Traditional crop yield prediction methods often rely on statistical models that are based on historical data and do not account for the complex interactions among environmental factors, soil, and crop genetics. Given the abundance of data accessible in agriculture, Machine learning algorithms can analyse large and heterogeneous datasets and identify patterns and relationships that traditional methods cannot capture. The main objectives of the review include: to explore machine learning and deep learning techniques used to predict the crop yield using various input parameters, to evaluate the accuracy measures considered for measuring the performance of these techniques, to examine the efficiency of these techniques, to explore the input parameters used for modelling, to compare the performance of algorithms for crop yield prediction and to explore the efficiency of hybridized models. In the current study, we performed a systematic literature review (SLR) to get insight of algorithms, input parameters and evaluation parameters used in the relevant papers. The target crops for the study are corn/maize and soyabean. We were able to identify 1859 related papers from four databases, out of which we selected 82 papers for further analysis after considering inclusion and exclusion criteria. The results of SLR shows that the most popular machine learning algorithm among researchers of selected domain are Random forest, Artificial Neural Networks (ANN), Support Vector Machines and XG Boost. Additionally, LSTM and CNN are the most widely used deep learning techniques for crop prediction. From the literature reviewed, it has also been shown that temperature, precipitation, crop historical yield, NDVI and ph-value are the most considered input parameters by the researchers of the selected domain. The current review has also shown the inclination towards the hybridization of models for enhancing the model accuracy.