Performance Evaluation of Best Feature Subsets for Crop Yield Prediction Using Machine Learning Algorithms

Abstract

The rapid innovations and liberalized market economy in agriculture demand accuracy in Crop Yield Prediction (CYP). In accurate prediction, machine learning (ML) algorithms and the selected features play a major role. The performance of any ML algorithm may improve with the utilization of a distinct set of features in the same training dataset. This research work evaluates the most needed features for accurate CYP. The ML algorithms, namely, Artificial Neural Network, Support Vector Regression, K-Nearest Neighbour and Random Forest (RF) are proposed for better accuracy. Agricultural dataset consists of 745 instances; 70% of data are randomly selected and are used to train the model and 30% are used for testing the model to assess the predictive ability. The results show that the RF algorithm reaches the highest accuracy by means of its error analysis values for all the distinct feature subsets using the same training agricultural data.