Crop Yield Prediction For Maize Using Machine Learning Case Study Asili Farms Kiryandongo District
Abstract
Agricultural production is an outstanding source of income for Ugandans, especially small-scale farmers, and has contributed to their daily living. However, due to certain seasonal changes attributed to climatic conditions, and socio-economic conditions among others, these farmers find themselves in a crisis of getting imprecise yields estimates out of the planting seasons. Traditionally, methods of yield estimation like the crop cut method are used to estimate yield which has led to imprecise yield estimates carried out by the farmers. Sampling techniques to determine yield by statistical methods is one of the methods that has been employed so as to improve on yield estimates. In this study, we investigate the use of machine learning algorithms for maize yield prediction with the goal of comparing maize yield estimates from a random forest model and a multivariate linear regression model based on Sentinel 2A NDVI, Landsat 8 LST, and rainfall data for six different years. From the statistical analysis Random Forest model performed better compared to Multivariate linear model making it to be a better model in the estimation of maize yield. This had a smaller root mean square error, mean absolute error, 𝑅2 and a lower PEMY of 19.6209, 9.6934, 0.861184 and 0.17177 respectively in comparison with those for the multivariate linear model which were 24.7182, 14.7698, 0.713861, and 1.627083 respectively implying that Random Forest can estimate Maize yields with better precision. In conclusion, since both methods yielded minimal percentage errors of maize yields 0.171766332 and 1.627083217%, respectively they can both be used to estimate maize yields.
Key words: Machine learning, Random Forest, multivariate linear regression, NDVI, crop yield