ACO-ELM: An Ant Colony Optimization-Driven Extreme Learning Machine for Dry Bean Variety Classification

Accurate classification of dry bean varieties is important for automated agricultural inspection, quality assessment, and efficient crop management. This study proposes an Ant Colony Optimization-based Extreme Learning Machine (ACO-ELM) for the classification of dry bean varieties using morphological characteristics. The Dry Bean dataset initially contained 13,611 samples with 16 numerical features representing geometric and shape-related characteristics across seven bean classes. To minimize potential data leakage, 68 duplicate records were identified and removed before data partitioning, resulting in 13,543 unique samples. The dataset was divided using a stratified 80:20 train-test split, producing 10,834 training samples and 2,709 independent test samples. An internal validation subset was further created from the training data for ACO-based hyperparameter optimization. ACO was employed to optimize the number of hidden neurons, regularization parameter, and activation function of the ELM using 20 ants over 20 iterations. The optimization selected 900 hidden neurons, a regularization parameter (C = 50.0), and a sigmoid activation function, achieving a validation accuracy of 94.83%. The optimized ELM was subsequently retrained using the complete training set and evaluated on the independent test set. The proposed ACO-ELM achieved 92.47% accuracy, 92.47% weighted precision, 92.47% weighted recall, and 92.46% weighted F1-score. Furthermore, the model achieved a 99.05% weighted multiclass ROC-AUC, with macro precision, recall, and F1-score of 93.65%, 93.45%, and 93.54%, respectively. The results demonstrate that ACO-based hyperparameter optimization can provide an effective ELM-based framework for multiclass dry bean variety classification.