Artificial Neural Networks Models for the Classification of Genotypes for Rice Yield

Abstract

Artificial neural networks, or ANNs, are widely used in many scientific fields like classification, image processing, identification, and prediction. 
This research was done at the SKUAST Jammu for classification of 140 rice varieties on the basis of their yield. Classification was made in terms of 
9 morphological characters such as yield per plant, number of days for 50percent flowering, number of days for full flowering, plant height, number 
of effective tillers per plant, panicle length, grain length, grain width and ratio of grain length & grain width. The yield per plant was grouped into 
three categories and used as the response variable, while the remaining traits served as predictor variables. The rice genotypes were then classified 
according to neural networks (NN) methods Multilayer Perceptron (MLP) and Radial Basis Function (RBF). The ability measures of classification 
such as Accuracy Rate, Kappa Statistics, Average Precision, Average Recall and F1 Score were used for testing samples. It is observed that RBF NN 
performed better than MLP NN for different classes of yield on the basis of classification ability measures. The variable effective number of tillers per 
plant found out to be important variable as per RBF NN whereas plant height was important variable in case of MLP NN.