Application of Machine Learning Techniques in Predicting Productivity of Sorghum

Abstract

In agriculture sector, machine learning techniques play an important role for predicting productivity of agricultural products. Some of applications of machine learning in agriculture is mentioned in Appendix B. Existence of yield constraints in sorghum production decreases productivity. Drought, insect, bird, disease, inadequate improved sorghum varieties are some of constraints found in sorghum production. To address these problems after trail, correlation and regression models are used in the sample data of sorghum. To answer research question of identifying factors that increase and decrease productivity of sorghum from sorghum sample data, Pearson correlation is used. Pearson correlation shows linear relationship between dependent (yield per a plot) and independent variables (replication, varieties, heads per a plot, drought score, emergency day, flowering day, maturity day, bird score, disease score, insect score and plant height). Positive correlation means if independent variable increases the dependent variable also increases while negative correlation means if independent variable increase then dependent variable will decrease. Correlation of yield per a plot with replication, varieties, heads per a plot and plant height shows positive correlation. And correlation of yield per a plot with drought score, emergency day, flowering day, maturity day, bird score, disease score and insect score tell negative correlation in RStudio. To answer how machine learning can be effectively used to predict productivity of sorghum, ridge regression, LASSO and elastic net regression are used in RStudio. Ridge regression describe variables with sum of squares of coefficient while least absolute shrinkage and selection operator define variables with sum of absolute values of coefficient and elastic net regression combines ridge regression and least absolute shrinkage and selection operator. Ridge regression, LASSO and Elastic net regression of root mean square deviation (RMSE) and mean square error (MSE) in RStudio indicates importance of variable as percentage value. Ridge regression RMSE variable are maturity day (100%), flowering day (79.5%) and drought score (32.9%) and MSE variables are maturity day (100%), flowering day (79.5%) and drought score (32.9%). For LASSO RMSE variables are maturity day (100%), flowering day (76%) and drought score (35.2%) and MSE variables are maturity day (100%), flowering day (76%) and drought score (35.2%). Elastic net regression RMSE variables are drought score (100%), maturity day (91.3%)

XV

and flowering day (69.8%) and MSE variables drought score (100%), maturity day (91.5%) and flowering day (69.9%). Elastic net regression is selected than ridge and LASSO as it has better features than others. Drought score, maturity day and flowering day are selected as most important factors used in machine learning for predicting productivity of sorghum. Therefore, machine learning is used to address a problem by applying different techniques in problem related data. In agriculture research centers, sorghum production constraints can be identified from historical data by applying different techniques including machine learning to problem related data.