Comparison of Soil Fertility Classification Algorithms Using a Data Mining Tool
R. S. PARMAR
College of Agricultural Information Technology, AAU, Anand, India.
H. K. PATEL
Main Forage Research Station, AAU, Anand, India.
G. J. KAMANI *
College of Agricultural Engineering & Technology, AAU, Godhra, India.
A. T. AGJA
International Agribusiness Management Institute, AAU, Anand, India.
*Author to whom correspondence should be addressed.
Abstract
To search for the possibility of soil fertility classification, a data mining technique was followed. The open-source data mining toolkit Weka was used for the development of a classification model for soil fertility. Thus, after cleaning and sorting the soil dataset, six classification algorithms, viz., Neural Network, Support Vector Machines, KNN, Random Forest, Random Tree, and Naive Bayes, were used over the soil dataset. The results indicated that the Bayesian-based algorithm has better performance than the function-based, lazy-based, and tree-based algorithms. The Naive Bayes algorithm was found to be the best fit in the classification of soil fertility as it recorded a classification accuracy of 97 %, a sensitivity of 0.97, and a precision of 0.97 as compared with other algorithms. The Naive Bayes algorithm has also achieved the highest F1 score of 0.97 and MCC of 0.94.
Keywords: Soil fertility, neural network, SVM, KNN, RF, random tree, naive bayes