Classification of Toothbrush-Handle Evidence Using Differential Raman Spectroscopy Combined with Peak-Region-Guided Chemometrics

Toothbrush handles may be recovered as polymeric evidence in forensic investigations, but similar bulk compositions can hinder brand-level discrimination. Differential Raman spectroscopy was therefore combined with a peak-region-guided chemometric fusion network (PGCMFNet) for toothbrush-handle classification. Differential Raman spectra were acquired from 110 samples representing nine brands. Cropped spectra, first-derivative spectra, training-fold-selected peak regions, and chemometric priors were integrated through a sample-level gating mechanism. Performance was evaluated by nested repeated stratified cross-validation against four baseline methods. The baseline methods were principal component analysis followed by support vector machine classification (PCA-SVM), partial least-squares discriminant analysis (PLS-DA), random forest (RF), and a one-dimensional convolutional neural network (1D-CNN). PGCMFNet achieved an accuracy of 93.27% and a macro-averaged F1-score of 88.02%, exceeding the corresponding baseline values. Ablation analysis indicated that the spectral branches and chemometric priors contributed to classification performance. The selected intervals corresponded to characteristic polymer vibrational regions, supporting the spectroscopic plausibility of the extracted features. These findings demonstrate the potential of PGCMFNet for closed-set, brand-level classification of toothbrush-handle samples within the evaluated dataset.