Development and Validation of a Machine Learning-Based Model for Predicting Frailty Risk in Head and Neck Cancer Patients Undergoing Radiotherapy

2.5 Data Preprocessing Data integrity verification was conducted using SPSS and Jupyter Notebook. Multivariate missing data underwent multiple imputation , with categorical variable gaps filled via modal substitution and continuous variables via mean imputation. Anomaly detection implemented Isolation Forest methodology, replacing outliers with feature medians. Continuous predictors were normalized to [0,1] range through min-max scaling, while categorical features underwent label encoding transformations. 2.6 Feature Selection LASSO (Least Absolute Shrinkage and Selection Operator) regression was employed to mitigate model overfitting. Optimal regularization intensity (λ_min) was established through 10-fold cross-validation, retaining predictors exhibiting non-zero coefficients at this threshold. These features constituted model inputs, with frailty status as the dichotomous outcome. Notes: Approved by Ethics Committee of Affiliated Hospital of SWMU (KY2025104) All personally identifiable information has been removed Written informed consent was obtained from participants