A clinical diagnosis prediction model based on XGBoost-SHAP

Clinical diagnosis is typically a binary classification problem influenced by multiple indicators. Accurately predicting whether a patient is diseased based on various test indicators, and identifying the main influencing factors for targeted treatment, is a highly challenging task. To address issues such as high complexity, low model interpretability, and sensitivity to noise in traditional models, this study constructs an XGBoost-SHAP model based on data from 425 acute heart failure (AHF) patients and 430 healthy individuals. The model is used to predict the diagnosis of acute heart failure (AHF) and identify relevant indicators. This model combines the XGBoost algorithm with SHAP (SHapley Additive exPlanations) for interpreting the machine learning model in clinical diagnosis prediction. The approach first sets multiple hyperparameters for the XGBoost model and uses grid search to find the optimal hyperparameter combination. Then, cross-validation is applied to assess the model’s accuracy and stability by dividing the data into training and validation sets. Finally, SHAP is used to identify the main influencing factors in the prediction, explaining the diagnostic process. The combination of XGBoost and SHAP aids in clinical diagnosis. Experimental results show that the XGBoost-SHAP combined model outperforms other models by effectively handling large amounts of features and data, achieving high accuracy, handling missing values, running at high speed, and offering good scalability and flexibility. Additionally, it can provide insights into the main influencing factors and explain the diagnostic process, which significantly aids in clinical diagnosis and validates the feasibility of the proposed method.