ObjectiveTo construct an artificial intelligence (AI) model for predicting palliative care referral of cancer patients based on public judgment, and to provide reference for referring cancer patients to a palliative care service. MethodsA mixed method (qualitative and quantitative) approach was used to identify indicators for palliative care referral of cancer patients based on public judgment. Clinical and follow up data from 354 advanced cancer patients were retrospectively collected, and Random Forest algorithm was applied to develop the prediction model. Additionally, 75 terminal cancer patients were prospectively studied for external validation of the prediction model, and a web based calculator was developed based on the model results. ResultsThe prediction model for palliative care referral of cancer patients included 12 indicators:pain, malnutrition, respiratory distress, decreased consciousness level, limited self care ability, restricted mobility, ascites, edema, depression, defecation and urination difficulties, dysphagia, and difficult to heal wounds. The optimal timing for referral should be with an estimated life expectancy of less than three months. The area under the receiver operating characteristic curve (AUC) for the prediction model was 0.876, and the Brier score was 0.128. The AUC in the external validation sample was 0.831, and the Brier score was 0.095. ConclusionThe public judgment based prediction model for palliative care referral of cancer patients demonstrates satisfactory predictive performance and can serve as a rapid screening tool for palliative care referral.