Soybean Disease Detection with Feature Selection Using Stepwise Regression Algorithm: LVQ vs LVQ2

Abstract

ndonesia's soybean needs increase from year to year. But according to data from the Badan Pusat Statistik (BPS) the amount of national soybean productivity is still low, so the fulfillment of soybean needs is done by importing soybeans from several countries such as China, Ukraine, Canada, Malaysia, and the United States. Low soybean productivity is caused by several factors. One of the causes is disease. This study aims to create a soybean disease detection by applying Learning Vector Quantization 2 (LVQ2) neural network algorithm(ANN) and Stepwise Regression Algorithm attribute selection. The attribute variables used consisted of 35 symptoms of the disease in soybean crop data. The data used in this study is a soybean dataset taken from University of California Irvine Machine Learning Repository as much as 200 data. The distribution of training data and test data is done by the k-fold cross validation method with a value of k = 10. The result of the study shows that the best paramater use in lVQ2. The results showed that the best parameters in LVQ2 is learning rate (α) value of 0.3; epsilon 0.04; and maximum epoch 100. While the best attribute selection uses the parameter p to enter and p to remove of 0.15 which produces 17 selected attributes such as date, plant stand, precipitation, leaves, leaf spot halo, leaf spot margins, leafspot size, leaf mildew, stem canker, stem fungi, external decay, fruit pods, fruit spots, seeds, mold growth, seed discolor, roots. The best results in this study resulted in an accuracy of 90.5%, 9.5% error rate, 90.5% sensitivity, and 98.94% specificity