Advancement Through Research
The purpose of the research is to find a robust and efficient model for plant disease detection. Therefore, the current study proposes an enhanced-DTC (Decision Tree Classifier) approach for high dimensional imbalanced dataset in plant disease diagnosis. In this approach, instead of just using traditional decision tree algorithm, its capabilities are enhanced with Random Over (RO) sampling method for class balancing and three well-known feature selection techniques, i.e., Consistency (Cons), Correlation-based Feature Selection (CFS), and Random Forest Importance (RFI) filter for dimensionality reduction. The proposed methodology aims to enhance the performance of the five most commonly used decision tree algorithms, namely, C4.5, Classification and Regression Tree (CART), Bagging CART (Bag-CART), Partial Decision Tree (PART-DT), and Boosted C5.0 (B-C5.0). Results specify that the enhanced-DTC approach performs superior to the existing decision tree algorithms for the multiclass Soybean Large (SBL) dataset. It has been observed that the enhanced-DTC approach with both RFI and C4.5 method performed the best with an Accuracy (ACC) of 98.10% and Area Under Curve (AUC) of 97.79%. A real-time application of the proposed model can be used by the agricultural experts to take preventive measures in the most sensitive areas that are prone to a particular disease. Hence, timely intervention would help in reducing the loss in productivity of plants which will further benefit the global economy, agricultural production, and the food industry.
Plant Disease, High Dimensional, Imbalanced Dataset, Enhanced Decision Tree Approach, Feature Selection, Random Over Sampling
