Aim: Current prostate cancer (PCa) diagnostics are based on PSA level assessment as well as histological analysis of biopsies. However, both false negative and false positive diagnoses are still made....

The stability of miRNAs in biological fluids and their association with pathological conditions position them as promising biomarkers for non-invasive cancer diagnostics, including those for non-small cell lung cancer (NSCLC). However, variability in miRNA expression and technical aspects of quantitative reverse transcription PCR (RT-qPCR) require effective normalization to accurately assess miRNA levels and identify biological differences. Despite the availability of multiple normalization methods, there is no consensus on the optimal approach for circulating miRNA data. This study provides a systematic methodological benchmarking of five normalization strategies—pairwise normalization, Tres normalization, Quadro normalization, normalization to the overall mean, and normalization to the exclusive mean—applied to RT-qPCR data of 17 miRNAs from plasma EVs of NSCLC patients and healthy donors. The aim was to evaluate their impact on the performance of machine learning models. For each normalization method, Random Forest models were built. Models were evaluated using accuracy, F1 score, AUC-ROC, AUPRC, precision, sensitivity, specificity, MCC, and overfitting gap. Pairwise normalization consistently outperformed all other methods across most metrics, achieving the highest AUC-ROC and AUPRC and the lowest overfitting gap, with statistically significant differences compared to other methods. Pairwise normalization offers a computationally simple, high-performing option for such benchmarking applications.