Advanced machine learning approaches for predicting florasulam concentration in the presence of tio2 nanoparticles in aqueous medium

Abstract

This thesis presents a comprehensive study on the prediction of Florasulam concentration in aqueous solutions containing Titanium Dioxide (TiO2) nanoparticles using advanced machine learning techniques. Florasulam, a widely used herbicide, poses significant environmental risks due to its persistence in water systems, particularly when interacting with nanoparticles like TiO2. This research leverages several machine learning models, including CatBoost, Random Forest, XGBoost, and Support Vector Machine (SVM), to model and predict the behaviour of Florasulam under varying environmental conditions such as pH levels and UV exposure. The study begins with an extensive literature review, outlining the environmental impact of Florasulam, the role of TiO2 nanoparticles in photocatalysis, and the application of machine learning in environmental sciences. Key concepts such as density, pH, photocatalysis, and the Point of Zero Charge (PZC) of TiO2 are defined and discussed in relation to their influence on the degradation of Florasulam. Experimental data were collected from controlled laboratory experiments, where Florasulam was exposed to different concentrations of TiO2 nanoparticles across various pH levels and under both UV and sunlight conditions. This data was supplemented with synthetic data generated through statistical modelling to ensure robustness in the machine learning training process. The core of the thesis focuses on the development and evaluation of predictive models. Each model was trained and tested using a combination of original and synthetic data. The models were evaluated based on performance metrics such as Mean Squared Error (MSE), Mean Absolute Error (MAE), and R2 scores, as well as training time and computational efficiency. CatBoost and XGBoost emerged as the most accurate models, with CatBoost showing particular strength in handling categorical data and achieving high prediction accuracy. Random Forest, while slightly less accurate, was noted for its faster training times and robustness against overfitting. SVM, despite its longer training time, provided valuable insights into the non-linear interactions between Florasulam and TiO2 nanoparticles. The study also explores the effects of pH and TiO2 nanoparticle concentration on Florasulam degradation. It was found that higher pH levels and higher concentrations of TiO2 significantly accelerated the degradation process, highlighting the importance of environmental conditions in determining the fate of herbicides in natural water systems. This research demonstrates the efficacy of machine learning models in predicting environmental contaminants' behaviour, offering a powerful tool for environmental monitoring and management. The findings provide a foundation for future research, suggesting avenues for expanding the dataset, integrating real-time data, and exploring deep learning approaches to further enhance predictive accuracy.