XAI-LungCancer: an eXplainable AI-based deep learning framework for predicting lung cancer using MLP-mixer and SHAP-enhanced classifiers

Abstract Lung cancer remains one of the most prevalent and deadly forms of cancer worldwide, necessitating robust predictive models for early detection and diagnosis. In this study, we present a deep hybrid architecture for lung cancer prediction using demographic and clinical features from the Survey Lung Cancer dataset. The proposed framework integrates MLP-Mixer modules with a regression branch for continuous severity estimation and a classification branch for binary diagnosis. In addition, we enhance conventional XGBoost and Multilayer Perceptron (MLP) classifiers with SHAP (Shapley Additive exPlanations) to provide model-agnostic feature attributions and improve clinical interpretability. The dataset comprises 284 patient records with 16 features and is evaluated using a 70/30 train-test split and 5-fold cross-validation to assess robustness. Under this evaluation protocol, the MLP classifier achieves an accuracy of 98.39%, with a precision of 0.9836, a recall (sensitivity) of 0.9936, and an F1-score of 0.9917, indicating strong agreement between precision and recall. The XGBoost model attains a competitive accuracy of 96.77% on the same test set, supporting the generalizability of the framework across different classifier backbones. SHAP-based analyses provide intuitive visual explanations that quantify the contribution of each clinical and demographic feature to the final predictions, aligning with established medical reasoning. Overall, the proposed framework combines high predictive performance with transparent feature-level explanations, offering a promising decision-support tool for the early detection and diagnosis of lung cancer.