Attribute Selection for Detection of Soybean Plant Disease and Pests

Abstract

Soybean is one of Indonesia's main commodities that is widely used as a secondary food source because of its protein content. This article compares three attribute selection algorithms, namely Backward Elimination, Forward Selection, and Stepwise Regression with Learning Vector Quantization2 (LVQ2) classifier to detect soybean to avoidance the diseases and pests. Attribute selection is needed at the pre-processing phase of soybean disease data. By selecting relevant data attributes, it is expected that detection accuracy can be maximally generated with minimum computation. The selected attributes are then classified using the LVQ2 method which is a variation of the development of LVQ. LVQ2 has the ability to classify several diseases better than LVQ with the existence of two reference vectors for weight update. The experimental results show that the best parameter for feature selection are p 0.25, a-enter 0.095 and a-remove 0.095 which can reduce the attribute up to 20 attributes with LVQ2 classification accuracy reaching 91%. The results of this accuracy can be obtained through all three selection algorithms.