Akademik Veri Yönetim Sistemi
Journal of Water Process Engineering, cilt.77, 2025 (SCI-Expanded)
Bioremediation is a highly effective method for metal removal from water, but experimental procedures require time and high costs. This study aims to examine the best machine learning (ML) models for predicting the removal of cadmium (Cd) from aqueous solutions using wild and recombinant strains of Escherichia coli, as well as environmental conditions that affect the removal rate. Additionally, Cd bioremoval was estimated from scanning electron microscope images of bacteria by image processing with deep learning. Thirteen ML methods were used to predict the bioremoval capacity of strains, but only nine were successful. AdaBoost (R2: 0.97) and ExtraTrees (R2: 0.99) models achieved higher prediction performance analyses of wild and recombinant bacteria, respectively. The results show that the MSE values for the training, experimental, and validation sets are low and the R2 values are high, which means that the error between the predicted and actual values is quite small. Since the initial Cd concentration is the most effective factor on the removal rate, its estimation from images has become even more important. In image processing analysis, the MobileNetV3-Small model was found to be the most successful. Results obtained after training with MobileNetV3-Small showed precision values of 97–99 % and 94–98 %, recall rates of 97–99 % and 94–98 %, and the range of the top 5 F1-scores of 96–99 % and 94–98 % for wild and recombinant bacteria, respectively. This study demonstrates the effectiveness of data- and image-driven approaches in predicting Cd removal performance by bacteria from Cd-contaminated waters, providing a paradigm reference for future ML applications.