DYNAMIC UNCERTAINTY-AWARE BAYESIAN NEURAL MODEL FOR REAL-TIME WORKFORCE DISPLACEMENT RISK UNDER AI AUTOMATION

Abstract

The growing adoption of artificial intelligence (AI) in the workplace requires moving workforce displacement prediction in real-time and with uncertainty. This paper presents a probabilistic deep learning system that combines Bayesian Neural Nets (BNN) with Deterministic Deep Neural Networks (DNN), Monte Carlo (MC) Dropout, and a hybrid ensemble architecture to perform task-level displacement prediction. Empirical testing of 18,796 standardized task-level data indicates that all models converge to the same place, with BNN Gaussian NLL converging to -0.94 and MSE converging to approximately 0.056. The Bayesian model produced better results with a MSE = 0.0558, RMSE = 0.2361, MAE = 0.1902, and R 2 = 0.0670, whereas the Hybrid model marginally increased the results (MSE = 0.0556, R 2 = 0.0691). Calibrated probabilistic forecasting was confirmed based on the diagnostic analysis of heteroscedastic residual patterns and positive correlation between predictive uncertainty and absolute error. The error in the segmentation in terms of classes was the lowest in the Medium-risk occupations and the highest in the High-risk groups. The framework will be useful in uncertainty-sensitive and dynamic displacement predictions under AI-based labor transformation.