I don't think SHAP alone can provide a solid foundation for anomaly detection explanations in a thesis. More investigation into the treatment of PCA-anonymization is needed to make this a convincing argument. https://www.reddit.com/user/LeaveTrue7987