Differentiation of Toxic Plant Species Using UV–Visible Spectroscopy and Chemometric Techniques with Forensic Relevance

Toxic plants are frequently implicated in accidental and intentional poisonings, yet forensic identification is challenging when samples are degraded, powdered, or morphologically unrecognizable. This study investigates the capability of nontargeted UV–visible (UV–Vis) spectral fingerprinting combined with chemometric and machine-learning approaches to differentiate ten forensically important toxic plant species using methanolic extracts. UV–Vis spectra (200–800 nm) were acquired in triplicate and processed through a rigorous preprocessing workflow including baseline correction, detrending, Standard Normal Variate transformation, Savitzky–Golay smoothing, derivative analysis, and z-score normalization to enhance shape-based spectral features. Despite strong visual overlap, unsupervised principal component analysis (PCA) and hierarchical clustering analysis (HCA) revealed clear species-level discrimination, with chemically related taxa forming consistent clusters. Supervised classification models partial least squares discriminant analysis (PLS-DA), support vector machines (SVM), and k-nearest neighbours (KNN) demonstrated well-separated class boundaries in reduced-dimension space using a 70:30 training–test split. High cosine similarity values (0.95–1.00) confirmed that discrimination relies on subtle multivariate differences rather than simple spectral matching. Overall, the results demonstrate that UV–Vis spectral fingerprinting, embedded within a chemometrics-driven framework, offers a rapid, low-cost, and minimally destructive screening tool for forensic identification of toxic plant material, supporting its integration into tiered forensic toxicology workflows.