Interpretable Human Resource Management via Deep Gaussian Process Variational Inference: Uncertainty Modeling and Capability Evo

    In human resource management and human capital decision-making, traditional machine learning models often prioritize predictive accuracy while overlooking model uncertainty and the dynamism of organizational capabilities, thereby failing to provide decision-makers with reliable and transparent evidence. Current applications of artificial intelligence in HR also face challenges related to privacy protection, model interpretability, and insufficient employee trust. To address these issues, this paper proposes an interpretable human resource management framework that integrates deep Gaussian processes, employs variational inference to model uncertainty in dynamic organizations, and introduces a Dirichlet process mixture model to characterize capability evolution. The method maps employee features, organizational dynamic indicators, and capability distributions into the multilayer structure of a deep Gaussian process, leverages the reparameterization trick and doubly stochastic variational inference for efficient training, and achieves structural interpretability by computing effective kernel functions. Experimental results demonstrate that the proposed framework outperforms nine baseline methods on tasks such as employee attrition risk classification and performance improvement prediction, achieving significant gains in uncertainty estimation, dynamic capability mining, and training efficiency. The contributions of this paper are as follows: it proposes an interpretable deep Gaussian process model tailored to dynamic organizations; it develops an HR analytics approach that integrates uncertainty modeling, capability evolution, and data privacy protection; and it empirically validates the effectiveness of the approach, laying a foundation for trustworthy and transparent human resource decision-making in the future.