Intelligent Personalised Training Recommender Systems For Occupational Health Risk Mitigation In Pharmaceutical Industries

Abstract

Occupational health risks remain acute in pharmaceutical manufacturing, where complex processes and exposure to potent compounds demand targeted safety interventions. Traditional, one‑size‑fits‑all training frameworks often fail to accommodate individual vulnerabilities, role‑specific hazards and shifting risk profiles. This study presents an Intelligent Personalised Training Recommender System (IPTRS) that formulates training assignment as a multi‑label classification challenge, ingesting operator attributeshealth status, job function, exposure historyand delivering customised module recommendations. We benchmarked three stateoftheart architecturesTabNet, AutoInt and xDeepFMon a real‑world pharmaceutical dataset. TabNet achieved a subset accuracy of 85.4 per cent (micro‑AUC ≈ 0.998) with near‑perfect precision (≈ 0.999) and a recall of 0.922, demonstrating its conservative yet reliable baseline performance. Both AutoInt and xDeepFM attained flawless results (subset accuracy, F1‑scores and AUC = 1.0), highlighting their aptitude for modelling complex feature interactions, albeit with a cautionary note on potential overfitting in heterogeneous settings. These outcomes advocate a hybrid deployment strategyleveraging TabNet’s high‑precision recommendations alongside deep‑interaction models for exhaustive coverageunderpinned by continuous validation, adaptive thresholding and integration with real‑time biosignal and environmental feeds. Practical guidelines for industrial adoption emphasise dynamic content delivery, rare hazard detection and seamless alignment with existing occupational health and safety infrastructures.