Abstract
Crop water stress detection involves many features which are categorized as climatic, geographical and biological data of crops. Machine learning models with feature subset identification play a vital role in early detection of crop water stress. In the proposed work, utilizes 13 infield climate data such as soil moisture, temperature and relative humidity collected from cotton cropland using smart sensors along with metrological data obtained from the Indian Rice Research Institute (IRRI), Adudurai, Tamilnadu, India. This work deployed a light weight Artificial Neural Network (LANN) in conjunction with a feature subset identification using Archimedes’ Optimization Algorithm (AOA). AOA is a nature-inspired optimization algorithm that utilizes the physical laws of flexibility and balance to find optimal features from the dataset. ANN detects the crop stress by employing this feature subset identified by AOA. The proposed model was evaluated using Accuracy, Precision, Recall and F1-score and compared with conventional machine learning algorithms (Logistic Regression, SVM, Decision Tree, Random Forest and Naïve Bayes) and meta-heuristic selection methods (Genetic Algorithm, Particle Swam Optimization and Ant Colony Optimization). The proposed model achieves a classification accuracy of 94.5% when compared with meta-heuristic methods (ACO, PSO and GA) achieves 92.8%, 91.9% and 89.2% respectively. These findings indicate that AOA-based feature selection with a lightweight ANN provides a scalable solution for crop water stress detection.
Keywords: AOA Feature Identification, Crop Water Stress, LANN Classifier, Light Weight ANN, Water conservation.