A predictive framework for assessing naproxen-mediated changes in Nitzschia dubia fucoxanthin levels via machine learning-driven dielectric sensing.

Naproxen (Nap), a widely used non-steroidal anti-inflammatory drug, can disturb algal physiology under sufficiently high exposure conditions. In this study, a laboratory-scale machine learning (ML)-driven dielectric sensor framework was developed to estimate culture-level fucoxanthin (Fx) responses in the marine diatom Nitzschia dubia under acute Nap stress, using 48 averaged observations obtained from eight Nap concentrations over a 0-18 d cultivation period. A customized open-ended coplanar waveguide (CPW) probe measured dielectric responses (S11, ε'), over 30 kHz-3 GHz, with 735 MHz selected as the representative frequency for subsequent analysis. These descriptors were integrated with optical indices (OD686, chlorophyll-a, and Fx) and analyzed using multivariate statistics and regression models. The ML-driven dielectric approach indicated a concentration-dependent Nap response, with a low-dose stimulatory trend at ≤5 mg/L and inhibition at higher concentrations. At higher Nap levels, Nap correlated positively with S11 and negatively with ε', Fx, and OD686 (Pearson's r = 0.27, -0.36, -0.46, and -0.47, respectively). Correlation and principal component analysis (PCA) indicated strong coupling among algal biomass-related variables, Fx, and dielectric descriptors. Multiple linear regression (MLR) provided a baseline linking S11 and ε' to Fx. Support vector regression with a radial basis function kernel (SVR-RBF) captured nonlinear Nap-dielectric relationships; six-fold cross-validation achieved R2 = 0.913 with RMSE = 19.48 mg/L and MAE = 9.92 mg/L. 2D PDPs further supported the interpretability of the SVR model, suggesting that dielectric spectroscopy coupled with nonlinear regression has potential to non-destructive estimation of algal Fx responses under controlled pharmaceutical stress.