Machine Learning–Based Optimization of Phthalate Extraction Conditions Using HS‐INME‐GC/MS Data

ABSTRACT Optimization of extraction conditions in GC/MS analysis remains challenging because experimental variables can interact in complex and nonlinear ways. In this study, a machine learning based chemometric workflow was applied to optimize headspace in‐needle microextraction GC/MS (HS‐INME‐GC/MS) conditions for four phthalates: dimethyl phthalate (DMP), diethyl phthalate (DEP), dibutyl phthalate (DBP), and di(2‐ethylhexyl) phthalate (DEHP). A dataset comprising 75 experimental data points was used with saturation temperature, adsorption time, and desorption time as input variables. Ten regression models were evaluated, and five candidate models were further examined using feature importance, partial dependence, Pareto front analysis, desirability functions, and supporting robustness evaluation. Feature importance and partial dependence analyses showed that adsorption time and desorption time were generally more influential than saturation temperature. Among the candidate models, SVR and XGBoost were retained for the final confirmatory interpretation because their predicted conditions were close to the condition evaluated in the confirmatory experiments. Confirmatory experiments at 50.0°C saturation temperature, 50.0‐min adsorption time, and 5.0‐min desorption time showed good agreement between predicted and observed responses. These findings suggest that the proposed workflow provides a complementary chemometric approach for model comparison, response visualization, and balanced optimization of analytical extraction conditions.