SHAP-Interpretable Cross-Platform Machine Learning for Classifying High-Performing Influencers Using Ratio-Based Engagement Metrics

This study aims to develop and evaluate a Shapley Additive exPlanations (SHAP)-interpretable cross-platform machine learning framework for classifying high-performing influencer segments on TikTok and YouTube using ratio-based engagement metrics. The study addresses limitations of follower or subscriber-based evaluation by constructing a platform-stratified performance label and applying leakage-controlled feature selection to reduce circular prediction bias. A total of 2,000 influencer records, consisting of 1,000 TikTok and 1,000 YouTube influencers, were analyzed using Random Forest, Extreme Gradient Boosting (XGBoost), Categorical Boosting (CatBoost), and Support Vector Machine (SVM). Model performance was evaluated using accuracy, macro precision, macro recall, macro F1-score, and Receiver Operating Characteristic–Area Under the Curve (ROC-AUC), while SHAP was used to interpret feature contributions. The tuned CatBoost model produced the highest observed performance, with 0.9925 accuracy, 0.9904 macro F1-score, and 0.9995 ROC-AUC. Additional validation showed stable model ranking across alternative thresholds, while proxy leakage ablation and cross-platform transfer indicated that results should be interpreted as internal classification consistency rather than full platform-independent generalization. SHAP analysis showed that ratio-based engagement variables, particularly like ratio and engagement rate, contributed more strongly than raw audience size. The findings demonstrate that influencer performance can be represented as an interpretable