Abstrak

Microplastic pollution poses a major global environmental threat, demanding effective removal strategies. Coagulation is among the most practical methods due to its cost efficiency, simplicity, and high performance, with iron-based (Fe-based) coagulants showing particular environmental and operational advantages. However, integrated approaches combining statistical and machine learning optimization for different microplastic types and sizes remain limited. This study applied a hybrid Response Surface Methodology (RSM) and machine learning framework to optimize Fe-based coagulation for polyethylene terephthalate (PET), polyethylene (PE), and polypropylene (PP) microplastics of various sizes. A Box–Behnken design (15 runs per polymer) was used, totaling 135 experiments. Removal efficiency was quantified gravimetrically after floc separation and drying. The optimized process achieved a maximum removal efficiency of (94.9?±?0.2)%, comparable to many previous reports. RSM yielded the lowest mean prediction error (1.80%), surpassing Linear Regression (2.74%) and Artificial Neural Network (5.02%) models trained using k-fold cross-validation to mitigate overfitting. Coagulant dose was identified as the most influential variable, followed by polyacrylamide (PAM) dose and pH. These findings provide a robust, data-driven framework for optimizing microplastic coagulation and highlight key operational factors governing efficient removal.