Demonstrating the Power of SHAP Values in AI-Driven Classification of Marvel Characters
Ho-Woong Choi, Sardor Abdirayimov · Journal of Multimedia Information System · 2024
The transparency and interpretability of machine learning models have become paramount in the era of Explainable AI (XAI). This study leverages SHAP (SHapley Additive exPlanations) values to elucidate the decision-making process of an XGBoost classifier trained to distinguish between ‘good’ and ‘bad’ Marvel characters based on their skill sets. This study highlights the nuanced interpretability SHAP values provide, bridging the gap between complex AI models and the subjective domains of storytelling and character development. It underscores the need for balanced datasets and careful model training to mitigate inherent biases. These findings contribute to the field of XAI, demonstrating the potential of SHAP values in complex classification scenarios and underscoring their role in advancing AI transparency and trustworthiness.