| Abstract: | Pancreatic cancer is one of the cancers that are not easy to detect early due to the lack of obvious disease characteristics in the early stage, the tumor is mostly located in the posterior abdominal cavity, and the lack of early diagnosis tools. Therefore, when it is diagnosed, it is often approaching the late stage. In recent years, the studies of pancreatic cancer are mostly single-factor analysis and multi-factor analysis to summarize one or more risk factors, including past diseases, physical signs, family genes and long-term eating habits, etc., and for early evaluation models of pancreatic cancer is less. This study uses Logistic Regression (LR), Deep Neural Networks (DNN), ensemble voting learning (Voting), ensemble stacking learning (Stacking) and other methods to establish different models. Compare the performance between different models. It is hoped that based on the patient's past medical history, the high-risk group can be judged through the model, and whether it has a high probability of suffering from pancreatic cancer within one year, so that the public and doctors are aware of the risk of pancreatic cancer early. In this study, the best model is 19 factors LR’s model. The accuracy is 70%, the sensitivity is 70%, the specificity is 70%, and the AUC is 0.78. The contribution of this study is to use non-invasive factors to identify Chronic Kidney Disease, but it is a preliminary evaluation and ultimately requires doctors to diagnose. |
