Explainable AI In Complex Learning Systems: Trade-Offs Between Transparency And Performance

Abstract

In high-stakes fields like human resource management (HRM), Explainable Artificial Intelligence (XAI) has emerged as a crucial strategy for resolving the opacity of intricate machine learning (ML) systems. This research examines the trade-off between predictive performance and model transparency in complex learning environments. Using the Explainable HR Attrition Dataset, which includes demographic, organizational, performance, compensation, engagement, and temporal attributes, the research models employee turnover behavior with high dimensionality. Research empirically compares traditional interpretable models with an advanced ML approach, the Improved White Shark Optimized Cascaded Random Forest (IWSO-CRF). Post-hoc explainability resources like the Shapley Additive Explanations (SHAP) model are incorporated to balance interpretability and accuracy, offering both local (employee-level) and global (organizational-level) insights into model decisions. To guarantee data consistency, missing values are handled using the proper imputation techniques. Principal Component Analysis (PCA) is also used for feature extraction, which minimizes multicollinearity and reduces dimensionality while maintaining important information. Results indicate that the proposed approach enhances interpretability without significantly sacrificing predictive performance, although challenges related to explanation consistency and fidelity remain. The IWSO-CRF model achieves superior results, with 94.8% accuracy, 0.94 F1-score, 0.95 precision, and 0.95 recall, outperforming baseline interpretable models. The findings also reveal that transparency-by-design models, such as linear approaches, often fail to capture nonlinear and high-dimensional patterns in HR data. Therefore, the research combining inductive ML techniques with deductive analytical methods to develop robust, interpretable, and high-performing AI systems.