Enhanced Artificial Neural Network Models Using Principal Component Analysis for Plants multiclass classification

In recent days, multi-class classification problem is a major challenge in many real world applications, especially in the field of Agriculture. The recent research work shows that the Artificial neural network works effectively in multi class problem. To make the neural network learn faster, one of the neural network parameters called learning rate is used as a constant value. In this proposed work, the value of the learning rate gradually decreased by a fixed constant value once in a fixed number of epochs, so that the network can learn better than the existing methods. The proposed method has been tested on two different neural networks such as backpropagation neural network and Chebyshev neural network with two different datasets for leaf classification and identification type of wheat. Dimensionality reduction techniques such as principal Component Analysis and Linear Discriminant Analysis have been employed to increase the accuracy of the prediction. The results reveal that the PCA based network produces better accuracy when compared to Linear Discriminant Analysis based Network. Compared to raw data, the PCA components of the data increases the reliability of the data and it is proved by the r-square measure, the coefficient which is used to measure how the data is close to fit the regression line. Finding the best statistical hypothesis test is a challenging problem for interpreting the results. In modern machine learning, especially in deep learning the statistical test to compare the models on a single test dataset is very important. In this paper, Mc Nemar’s test is employed to show the significance of the result and it has been proved as a suitable test for evaluating the models.