Predict and Act: Machine Learning Driven Length of Stay Forecasting for Hospital Capacity Management in 2025 India

Predict and Act presents a machine learning driven framework for forecasting hospital length of stay (LoS) and translating predictions into real-world capacity gains in India’s strained 2025 healthcare system. Leveraging a publicly available dataset (318 438 admissions) and two off-the-shelf classifiers (logistic regression and k-nearest neighbors), we achieved a macro-AUROC of ~0.71 and macro-F1 of ~0.36 on a temporal hold-out. Embedding these models into admission workflows and daily capacity huddles produced an estimated 3 153 bed-days saved per month equivalent to freeing ~56 beds and avoiding ₹6.2 million in occupancy costs. Framed through dynamic-capabilities theory, our three-step playbook (secure data integration, bedside risk scoring, and multidisciplinary scheduling rituals) demonstrates how “sensing” and “seizing” routines operationalize LoS forecasts without compromising equity or regulatory compliance. Contributions include (1) quantifying the financial and operational value of predictive analytics beyond accuracy metrics, (2) extending dynamic-capabilities scholarship to frontline care, and (3) offering a replicable template for Indian hospitals to augment capacity by 7–10% amid chronic bed shortages.