TabFuseNet: A Cancer Prediction Model Utilizing Sample Cohort Data

Cancer is one of the leading causes of death worldwide, and early diagnosis can significantly improve patient survival rates and quality of life. However, traditional cancer diagnostic methods are costly and time-consuming, posing limitations on early detection. This study aims to develop an AI model that predicts the likelihood of cancer occurrence within five years based on patients’ health screening data, medical history, and prescription records. To achieve this, we utilized the Sample Cohort 1.0 dataset provided by the National Health Insurance Service (NHISS) and used health check-up and prescription data from 2007–2008 to predict the occurrence of lung, gastric, breast, liver, and colorectal cancer between 2009 and 2013. The dataset used in this study is structured in tabular format, and we employed machine learning models (SVM, Random Forest, XGBoost, LightGBM) as well as deep learning models (CNN, TabTransformer, TabNet) for performance comparison. The models were evaluated using AUC, F1-Score, AUPRC, and ACC. Finally, we propose the TabFuseNet model. TabFuseNet achieved an average AUPRC of 0.9248, demonstrating superior performance compared to conventional machine learning and deep learning models. Notably, TabFuseNet outperformed TabTransformer in handling tabular datasets, indicating its potential for applications in various research fields.