Raman Spectroscopy Coupled With Chemometrics and Machine Learning for Exploratory Property Prediction in Polypropylene Carbon Nanotube Nanocomposites

Raman spectroscopy was coupled with chemometrics and machine learning to explore associations between spectral descriptors and functional properties of polypropylene/carbon nanotube (PP/CNT) nanocomposites containing 0–5 wt% CNT. The dataset comprised 105 spatial spectra nested within 21 independently compounded batch specimens across seven compositions, with each specimen‐specific property response assigned to its five corresponding Raman spectra. Spectra underwent artifact filtering, asymmetric least‐squares baseline correction, vector normalization, and Savitzky–Golay smoothing. Principal component analysis (PCA) was used for spectral exploration, while partial least squares (PLS), random forest (RF), and XGBoost were evaluated for property prediction. Under specimen‐grouped validation, XGBoost achieved R 2 /RMSE values of 0.96/0.29 for electrical conductivity, 0.94/0.10 for storage modulus, and 0.97/0.91 for EMI shielding effectiveness. Under leave‐one‐composition‐out (LOCO) validation, the corresponding values were 0.81/0.52, 0.85/0.15, and 0.88/1.42, respectively. The results support exploratory within‐series property prediction and show the expected loss of accuracy when transferring to an unseen CNT loading. SHAP analysis was used for model‐specific interpretation rather than causal inference.