A MACHINE LEARNING MODEL FOR PULSES CROP YIELD PREDICTION: THE CASE OF HADIYA ZONE

Abstract

Agriculture in Ethiopia is the fundamental engine of its economy and employs the majority of the Ethiopian people. Most of these are smallholder farmers’ practice subsistence farming. These farmers whose output is the production of pulses such as Faba bean, Haricot bean and Peas which contributes to smallholder income as a higher value crop than cereal crops in Ethiopia. Pulses crop yield primarily depends on climatic conditions, diseases and pests, geographical and biological factors and the likes; these results decrease in the crop production. Currently, prediction of pulses crop yield is performed by the farmer’s based on long-term experience through visiting the condition on a particular field and using traditional statistical analysis. However these methods are subjective, insufficient ground observation, substantial inaccuracies might occur, resulting in inaccurate prediction of pulses crop yield. Thus, this research introduces the development of different predictive models using supervised machine learning techniques to predict the future pulses crop yield that to solve the aforementioned problems. Predicting pulses crop yields early is critical in order to plan and make various policy decisions like import-export, storage, pricing and marketing. The aim of this study isto develop a predictive model for pulses crop yield using supervised machine learning techniques in the case of Hadiya Zone. The researchers used Random Forest, Extreme Gradient Boosting, Decision Tree, K-Nearest Neighbor and Polynomial Regression algorithms. And also data analysis and implementation is done by using Anaconda software tool which consists of Python IDLE, Jupyter Notebook and Spyder. The performance of these models were evaluated by using different performance metrics like R-square, mean squared error and cross validation mechanism. The R square achieved in these five models was compared and Random Forest model is the best predictive model with R square of the 0.9711 and mean square error of the 0.4126 than the rest of the aforementioned models. Therefore, Random Forest Regression model is best outperformed and effective model for pulses crop yield prediction. Promisingly, this model will be used by the agricultural sector to have effective decision and policy making practices