Phase polyphenism in the Desert locust underpins swarm formation and represents a major challenge for early detection and control. Existing approaches for assessing phase state rely on behavioral assays or morphological traits that are either impractical in the field or lag behind rapid physiological change. Here, we demonstrate that resonance Raman spectroscopy of hemolymph, coupled with machine learning, enables rapid and minimally invasive discrimination of density-dependent phase state. Using 488 nm excitation, we show that visually similar Raman spectra contain sufficient multivariate structure to classify long-term solitarious and gregarious individuals with >90% macro F1 (harmonic mean of global precision and recall). Comparable performance between linear and nonlinear models indicates that phase-associated biochemical variation is captured through relatively direct spectral relationships, while poor classification of sex (~60% accuracy) confirms that these signatures are largely sex-independent. Time-course experiments further reveal asymmetric transition dynamics, with rapid spectral shifts following isolation and delayed responses to crowding, suggesting a lag between behavioral and biochemical phase change. Together, these findings establish Raman-based metabolic profiling as a promising platform for real-time monitoring of locust phase state and provide new insight into the temporal dynamics of density-dependent physiological transformation.

The CvI datasets contain the fully isolated and crowded S. gregaria juveniles' Raman spectra of hemolymph. The IandG datasets contain the time-dependent transition study juvenile Raman spectra for phase state prediction.