Detecting melanoma in National Health Interview Survey respondents using machine learning

Summary: Abstract Body: Melanoma represents the 5th most common cancer in the United States and contributed to over 8,000 cancer deaths nationwide in 2024. Early detection may prevent melanoma metastasis and death. Machine learning (ML) has shown to be successful in the early diagnosis of skin cancer. ML is both an accessible and inexpensive clinical implementation. This study aims to develop an ML model that detects melanoma in respondents from a U.S.-based national survey. Using the 2023 National Health Interview Survey, this study developed an XGBoost-based ML model to predict melanoma. The dataset included 29,522 adult respondents. Preprocessing involved imputation, categorization, and data leakage and class imbalance mitigation. Feature selection employed Spearman correlation analysis to identify significant associations while variables with low variance were removed to prevent data leakage. Moderate predictive performance was observed with an accuracy of 81.25% and ROC-AUC of 0.7932 using our ML model. For the majority class (no melanoma), it achieved a specificity of 81.91%, showing the model’s ability to correctly identify negative cases. For the minority class (melanoma), the model yielded a precision of 17.31%, a recall of 69.23%, and an F1-score of 27.69%. The model’s balanced accuracy was 75.57%. Cancer-related variables, such as a history of breast cancer and skin cancer, along with demographic factors like Hispanic identity, were observed to be the strongest predictors of melanoma. The ML model showed strong potential for integration with survey responses at a national scale. Therefore, this model has promise in detecting melanoma risk and supporting early screening efforts to reduce the burden of cancer, relying solely on survey data. Dany Alkurdi¹, Lara Shqair¹, Saniya Tariq², Ezdean Alkurdi³, Omar Alani¹, Dev Patel¹, Zachary Schwager² 1. Icahn School of Medicine at Mount Sinai, New York, NY, United States. 2. Lahey Hospital & Medical Center, Burlington, MA, United States. 3. University of Massachusetts Chan Medical School, Worcester, MA, United States. Pigmentation, Melanoma, and Melanoma Immune Surveillance