Application of extreme learning machine in plant disease prediction for highly imbalanced dataset

Abstract

Plant diseases are responsible for global economic losses due to degradation in the quality and productivity of plants. Therefore, plant disease prediction has become an essential area of research for agricultural scientists. The current study implements Extreme Learning Machine (ELM) algorithm for plant disease prediction based on a dataset collected in real time scenario namely Tomato Powdery Mildew Disease (TPMD) dataset. Since, the collected TPMD dataset was imbalanced thus; various resampling techniques namely Importance Sampling (IMPS), Synthetic Minority Over-sampling Technique (SMOTE), Random under Sampling (RUS), and Random over Sampling (ROS) have been used here for balancing the dataset before using it in the specified prediction model. ELM models have been developed for each of the balanced TPMD datasets obtained from these resampling techniques as well as for the imbalanced TPMD dataset. Area under Curve (AUC) and Classification Accuracy (CA) are considered as performance measures to analyze the performance of ELM model. Results show that ELM performed better for TPMD dataset using resampling techniques. Best results are obtained with IMPS technique having maximum values for CA and AUC i.e. 89.19% and 88.57% respectively.