ABSTRACT
Wilt disease detection is one of the major issue of remote sensing. Here, we have used wilt dataset available at University of California Irvine (UCI) Repository that includes five features of multispectral images acquired by Quickbird satellite sensor. The purpose of the study is to advance the efficiency and performance of the algorithm for wilt disease detection system. In this paper, we have developed a method for wilt disease detection system. The proposed algorithm involves two steps: (1) the preprocessing of dataset that includes standardization of dataset using z-score normalization and oversampling of minority-class training data using Synthetic Minority Over-Sampling Technique (SMOTE), and (2) classification of data into wilt or normal class using deep learning feed-forward networks designed with H2O Automated Machine Learning (AutoML). The performance of dataset is measured in terms of accuracy and area under curve (AUC). The accuracy and AUC of the proposed method are obtained as 88.8 percent and 0.9553 respectively, which are quite higher in comparison to Support Vector 510Machine (SVM), Extreme Gradient Boosting (XGBoost), and Deep feed-forward networks algorithms.
