Predictive modeling of medical waste and a proposal to improve segregation in a peruvian hospital

Accurate estimation of medical waste generation is essential to improve segregation practices, reduce infection risks, and enhance hospital sustainability, particularly in resource-limited settings. This study aimed to predict waste generation in a hospital in Amazonas, Peru, and to propose improvements in waste segregation using the Deming (PDCA) cycle. A retrospective study was conducted using monthly data from 2017 to 2024. Waste was classified according to Peruvian regulations and analyzed using autoregressive integrated moving average (ARIMA) time series models. Stationarity was assessed using the augmented Dickey-Fuller test, while seasonality was examined through autocorrelation and partial autocorrelation plots, as well as seasonal trend decomposition based on Loess (STL); these plots were also used to identify model orders. Model validation included information criteria, the Ljung-Box test, and error metrics. The selected models were ARIMA(2,1,2) for special waste (AIC = 999.58; BIC = 1014.90; RMSE = 43.33; MAPE = 65.16%), ARIMA(1,1,0) for biohazardous waste (AIC = 1291.01; BIC = 1298.67; RMSE = 208.17; MAPE = 22.78%), and ARIMA(1,2,1) for general waste (AIC = 302.97; BIC = 307.54; RMSE = 17.60; MAPE = 2.51%). Projections for 2025-2026 indicate a continuous increase in waste generation. The proposed intervention integrates educational and behavioral modification strategies, providing a methodological framework to support improvements in waste segregation and hospital planning.