Prochloraz (PRO) is commonly used to maintain freshness during the postharvest storage of citrus. However, excessive residues are detrimental to human health. This study integrated surface-enhanced Raman spectroscopy (SERS) with a characteristic optimization strategy to develop a rapid detection method for PRO residues in citrus. The prepared gold nanoparticles (AuNPs) demonstrated excellent enhancement effects, stability, and reproducibility. Firstly, four characteristic optimization algorithms were employed to identify the effective characteristic peaks of PRO from the entire SERS spectrum. Linear partial least squares (PLS) and nonlinear support vector machine (SVM) models were constructed. The performance comparison results revealed that the constructed PLS and SVM models based on the characteristic peak screening methods demonstrated better predictive performance than those using a single feature peak. More notably, LASSO displayed outstanding optimization capabilities, only selecting 11 characteristic peaks, representing 1.46 % of the full SERS, to develop the optimal SVM model with superior predictive ability while simplifying model redundancy. The developed method achieved a minimum detection concentration of 0.012 mg/kg, and a single sample was completed within 15 min. Anti-interference experiments and gas chromatography (GC) validated the significant anti-interference capability, the accuracy, and reliability of the method, satisfying the actual detection requirements for PRO in citrus.