An explainable hybrid AI architecture for operations research in industry 4.0: deep learning and interpretable modelling for intelligent decision support

This study proposes an Extended Deep Neural Network-Logistic Regression (EDNN-LR) architecture as a predictive-prescriptive decision-support model for Management Science (MS) and Operations Research (OR) applications in Industry 4.0 environments. The extended architecture integrates convolution-pooling layers, a Conditional Random Field (CRF) module, and a logistic-regression calibration head to improve convergence stability, interpretability, and optimisation performance. The model was evaluated through two complementary experiments using data derived from CoinMarketCap (2013−2021). The first experiment, conducted in MATLAB R2024a on a high-liquidity subset (AVAX-BTC-ETH), assessed predictive performance and achieved 98% classification accuracy with an AUC of 0.846. The second experiment, implemented in Python 3.12 using TensorFlow 2.17, evaluated computational scalability on the complete dataset comprising approximately 4.44 million records across 8,927 cryptocurrencies. Although large-scale discrimination performance was limited (AUC ≈ 0.50), the model demonstrated stable convergence and efficient processing of highly heterogeneous market data. Comparative analyses indicate that EDNN-LR converges faster than a baseline Deep Neural Network (DNN) while preserving interpretability through its logistic-regression component. A reproducible software implementation developed by the corresponding author accompanies the study and is documented in a supplementary Software Impacts article. The findings demonstrate the potential of EDNN-LR as an explainable decision-support framework for forecasting and optimisation under complex, data-intensive environments, while cross-domain applications in logistics, energy, and supply-chain systems remain a direction for future research.