Cutaneous melanoma is an aggressive and molecularly heterogeneous malignancy in which metastatic progression is associated with clinically relevant changes in tumor behavior, treatment response, and patient outcome. Although RNA-sequencing data provide a valuable resource for discovering molecular candidates, the high dimensionality of transcriptomic profiles and the limited sample size of many public cohorts can make single-model biomarker selection unstable and difficult to interpret.

This study developed a consensus multi-algorithm framework to classify primary and metastatic cutaneous melanoma and to prioritize reproducible transcriptomic candidate biomarkers using TCGA-SKCM RNA-sequencing expression data retrieved through cBioPortal. The analyzed cohort included 90 samples, comprising 70 primary cutaneous melanoma and 20 metastatic melanoma cases profiled as normalized RSEM z-score expression values.

More than twenty supervised, unsupervised, and statistical learning approaches were evaluated within a leakage-controlled cross-validation workflow. Preprocessing, feature selection, model optimization, and feature-importance estimation were performed within training folds, followed by validation on held-out folds. Candidate genes were ranked using a consensus score integrating selection frequency, normalized model-based importance, and cross-validation stability.

Discriminant Analysis achieved the best classification performance (Accuracy = 0.979, Precision = 0.981, Recall = 0.977, F1-Score = 0.979, AUC = 0.981), while Bayesian Network and TwoStep-based models showed consistent predictive behavior. Consensus integration prioritized FBXO7, PTGDR2, SARAF, WFDC10A, RPL13AP6, and RGS7 as stable candidate biomarkers.

Functional enrichment suggested that these candidates are linked to cancer-related pathways, including PI3K-Akt signaling, MAPK signaling, p53 signaling, cell-cycle regulation, apoptosis, and immune-response processes. Overall, the proposed framework provides a reproducible computational strategy for prioritizing melanoma-related transcriptomic candidates, while emphasizing the need for independent validation before clinical interpretation.