MACHINE LEARNING IN AGRICULTURE WITH APPLICATION IN MAIZE (Zea mays) YIELD PREDICTION MODELING IN UASIN GISHU COUNTY, KENYA

Abstract

Artificial intelligence is a subfield of computer science that aims to bring forth machines capable of emulating human behavior and replicating the cognitive and behavioral processes exhibited by humans. It is the discipline of making computers behave without explicit programming. The system functions by consolidating large amounts of data through efficient and repetitive processing, alongside the utilization of intelligent algorithms. This allows the software to independently acquire knowledge from patterns or attributes included in the data. Machine learning is a subfield of artificial intelligence that facilitates the autonomous acquisition of knowledge by computers through the analysis of past data, without the need for explicit programming. The primary objective of implementing machine learning techniques in the agricultural domain is to enhance both crop productivity and quality. This is motivated by the rise of big data technology and highperformance computation. It has propelled advancements in unraveling, quantifying, and comprehending data-intensive agricultural operational processes. The nature of machine learning models might vary between descriptive and predictive, depending on the specific research challenge and queries at hand. This study undertook a systematic literature review to assess the adoption of machine learning techniques in agricultural research in the Science Direct database to evaluate trends in its adoption, particularly in agricultural research. To evaluate machine learning applications in crop production, animal production, soil management, and agricultural mechanization, as well as the specific areas of study. Crop modeling and yield prediction is a decision tool used by farmers and other decision-makers in the agricultural sector to increase production efficiency and assist them in making swift decisions that affect the standard of agricultural output. Crop yield forecasting models can reasonably estimate the actual yield, but it would be preferable if they performed better. It is one of the most important precision agriculture topics. The need to adopt modern regression techniques of machine learning to attain sufficient amount of maize for sustainable agriculture, for food security, economic stability and nutritional benefits to the farmer. The study applied Random Forests, K Nearest Neighbor, and Extreme gradient boosting-XGBOOST machine learning regression algorithms to predict maize yield in Uasin Gishu county-Kenya using field-collected questionnaire data from 900 farmers spread across 30 wards in the five sub-counties. It utilized the R software and a train-totest ratio of 80:20. All the models could predict maize yield. Finally, model evaluation was done using Root Mean squared error-RMSE, Mean Squared Error-MSE, Mean Absolute Error-MAE, Mean Absolute Percentage Error-MAPE, Nash-Sutcliffe Efficiency Coefficient- NSE and Willmott's Index-WI to select the best model for yield prediction. Overall, XGBOOST emerged as the best regression algorithm in four evaluation metrics with RMSE of 0.4563, MSE =0.2082, MAE =0.3532, and Willmott’s index of 0.3264. XGBOOST was followed by Random Forest regression and K Nearest Neighbor regression algorithm. The findings recommend an XGBOOST machine learning regression model to predict maize yield in Uasin Gishu-Kenya to optimize maize yield for economic stability and food security. XGBOOST is an ensemble learning algorithm, there is a need to evaluate other ensemble regression algorithms from bagging and stacking in yield prediction and appreciate the need to fast implement machine learning techniques to make Agriculture more sustainable for future generations in Kenya