Assistant Professor Savannah State University, Georgia, United States
Abstract Submission: Predicting the uptake and translocation of organic contaminants in plants is crucial for assessing groundwater quality because plants play a significant role in the movement and fate of contaminants within ecosystems. Predicting the transpiration stream concentration factor (TSCF) and other concentration factors is essential in understanding the plant uptake of organic contaminants because these factors determine how contaminants are absorbed, distributed, and accumulated within the plant. The accurate prediction of the concentration factors, such as TSCF, facilitates the implementation of preventive measures and appropriate agricultural practices to reduce contamination risks. However, traditional mechanistic and numerical modeling methods often fail to reliably predict the TSCF as a criterion that explains uptake efficiency. This study developed a hybrid deep model to predict TSCF for organic contaminants by integrating convolutional neural networks (CNN) and long short-term memory (LSTM). The data was divided into training and test sets with ratios of 80% and 20%, respectively. The model used nine physicochemical properties of contaminants to predict TSCF. Random search cross-validation was applied to determine the optimized architecture hyperparameters for the hybrid CNN-LSTM model. The results demonstrated a successful training procedure for predicting TSCF by the model. The results indicated training and test losses for predicting TSCF were both in the same order and very close to zero. This study indicated that the hybrid model outperforms mechanistic models and has a higher performance compared to the classical machine learning models. Feature importance analysis highlighted the role of lipophilicity and molecular weight in predicting uptake and translocation of organic contaminants.
