Predictive modeling of microplastic adsorption in aquatic environments using advanced machine learning models.

This study addresses the critical environmental concerns surrounding microplastics, aiming to elucidate the intricate factors influencing their behavior and interactions with organic pollutants. Utilizing advanced artificial neural network modeling techniques, including GRU, LSTM, RNN, and CNN, a comprehensive analysis of microplastic sorption capacity and underlying mechanisms is conducted. The research relies on a meticulously curated dataset encompassing fundamental parameters such as organic compound composition, n-octanol/water partition coefficient, covalent acidity, covalent basicity, molecular polarizability to volume ratio, and the logarithm of the partition coefficient. Findings underscore the significance of understanding the n-octanol/water distribution coefficient (Log D) in predicting organic pollutant fate in aquatic environments, with compounds displaying higher Log D values exhibiting heightened affinity for microplastics, posing substantial ecological and human health risks. Additionally, the study highlights the importance of selecting appropriate models to accurately capture complex sorption processes, especially in varied aquatic environments. Furthermore, the profound impact of acidity, molecular polarizability to volume ratio, and covalent basicity on microplastic behavior is elucidated. Notably, machine learning models and CNNs demonstrate remarkable speed in prediction generation. A comparative analysis of four robust machine learning models establishes fundamental alignment between model predictions and empirical findings. Particularly noteworthy is the RNN model, emerging as the most accurate with an impressive accuracy of 0.967 and a minimum absolute error of 0.38. These results underscore the efficacy of the RNN model in predicting microplastic dynamics, holding promise for significant contributions to addressing microplastic pollution.