SHAP-Optimized Explainable Computational Intelligence Models for Interpretable Chemotherapy Adverse Event Prediction

Abstract

The negative effects of chemotherapy can be varied among patients, which will greatly reduce the continuity of the treatment and the clinical outcomes. The existing risk assessment approaches are mainly based on generalized statistical techniques or opaque models, which would tend to be susceptible to incomplete clinical evidence and would not deliver the explanations required to offer clinical faith. To address these limitations, this paper proposes an explicable artificial intelligence-based model of forecasting adverse extreme-type chemotherapy side effects using XG-Boost, random forest, and decision tree models, and using SHAP-based interpretability. The clinical, demographic, and treatment-related data were processed in an orderly manner, and the models that were trained using the data were processed with missing values. The experiments indicate that XG-Boost is more powerful in prediction and robust and generalization, as it has a reduced cross-validation error and improved confusion matrix performance with a reduced number of false negatives and false positives. Random forests and decision tree models proved to have higher misclassification errors, in particular, false negatives, and limit clinical validity. SHAP analysis provided patient-specific and globalized explanations of the risk factors that made predictions. In general, the proposed explainable AI architecture is better than the existing approaches regarding its capacity to be precise, resilient, and interpretable with reliable and tailored clinical decision support in cancer treatment.

Ähnliche Arbeiten