Chitosan-Based Flocculant Heavy Metal Removal Prediction Using Machine Learning Models

The expanding impact of heavy metals (HMs) on environmental
and
public health necessitates the development of advanced predictive
models that enhance the precision and efficiency of monitoring and
remediation strategies. This study aimed to evaluate newly developed
machine learning (ML) models for predicting the removal of HMs such
as cadmium (Cd2+), copper (Cu2+), nickel (Ni2+), lead (Pb2+), and zinc (Zn2+) using
chitosan-based flocculants (CBFs) from wastewater. A gradient boosting
regressor (GBR), Hist gradient boosting regressor (HGBR), random forest
regressor (RFR), and extreme gradient boosting regressor (XGBR) were
developed, with a cluster label generated by K-means clustering included
as an additional feature to enhance model learning. The ML models
were built using experimental data sets of HM ion removal across 484
sets of flocculation experiments involving various ions of HMs such
as Cu2+, Pb2+, Cd2+, Zn2+, and Ni2+. Results indicated that the HGBR model revealed
higher performance in combined HM removal scenarios, achieving a determination
coefficient (R2 = 0.94/0.97 for the testing/training
phases. For individual metals, all models achieved excellent accuracies,
especially for nickel (Ni2+), with the GBR model obtaining
the lowest error rate in the testing. The results signified a robust
capability of the HGBR model for generalization and its capacity as
a trustworthy tool in the framework of environmental monitoring. Future
research directions required the exploration of the synthesis of these
models into real-time predictive monitoring systems and an exploration
of the application of integrated ML approaches to boost the predictive
accuracy and reliability across wider environmental conditions.