Fig. 1

Overview of the study workflow. (a) Study design and classification framework: Cervical scraping samples (n = 160) were processed using the Illumina Infinium MethylationEPIC BeadChip. The dataset was split into training data (n = 120) and testing data (n = 40). A two-step classification model was developed, where Step One classified samples as normal or tumor (benign + malignant), and Step Two further distinguished benign from malignant tumors. (b) Data preprocessing pipeline: Methylation data underwent quality control, normalization, and filtering to ensure high data quality. *Indicates the filtered criteria in each step. (c) Machine learning model development: Feature selection was performed using recursive feature elimination with cross-validation (RFECV), followed by model training. Multiple machine-learning models were evaluated for Step Two model, with gradient boosting selected as the final model for classifying benign and malignant tumors