Productivity and Soil Fertility Prediction Model Using Machine Learning: The Case of Southern Ethiopia

Abstract

Melese Z. Segaro, Abenezer Daniel and Habtamu S. Adugna

One of important part that is source of the nutrients to grow crops and for successful agricultural production is soil. Capability of soil to sustain crop growth by providing significant nutrients and favorable chemical, physical, and biological characteristics as a habitat is called soil fertility. There are different properties on which several types of crops grow. Characterizing soil property fertility level, which crops to sow in certain soil types is important for successful crop productivity. Southern Ethiopia is reach with fertile containing substantial minerals soil and major producer/supplier of fruits, vegetables, and cereals root crops but crop production less than demand. This research study is basically aimed at designing productivity and soil fertility prediction models using machine learning techniques. Researchers conducted different experiments using dataset containing 2200 soil fertility records of 22 well producing crops from Southern Ethiopia. There are five machine learning algorithms used for comparative analysis to select best performance one using K-fold cross validation. Such as Decision Trees(90%), Logistic Regression (95%), SVM(13%), Random Forest(99.5%), and Naive Bayes(99%) accuracy achievement in each of them. As a result, Random Forest algorithm was built with best performance with 99.5% accuracy and decided to design the model for successful prediction of soil fertility and productivity for particular crop in Southern Ethiopia.

PDF