Demand forecasting for recycled plastics from waste in multi-stage supply chain: A comparative study

The bullwhip effect is a persistent challenge in multi-stage recycling supply chains, where fluctuations in waste availability, recovery rates, and downstream demand amplify variability across upstream recycling processors. These distortions increase operational costs, destabilize production planning, and reduce the efficiency of recycled-plastic markets. Although prior research has explored forecasting techniques for traditional supply chains, limited attention has been given to forecasting material flows derived from waste streams, where uncertainty is considerably higher. To address this gap, this study proposes a data-driven forecasting framework that combines machine learning models with ARIMAX and neural networks to predict demand for recycled plastic products. By incorporating external causal factors, macroeconomic indicators, and time-series patterns, the models improve forecast precision for recycled material flows. A comparative evaluation against conventional forecasting methods demonstrates that machine learning-based approaches significantly reduce demand distortion and enhance supply-chain coordination in recycling-focused systems. The findings reveal that proposed models outperform traditional ARIMA-type techniques, achieving up to 94% forecasting accuracy, thereby improving inventory planning, reducing variability, and strengthening the recycling system’s resilience. These results highlight the potential of advanced data-driven forecasting to support circular-economy objectives by enabling more stable, efficient, and predictable recycled-plastic supply chains.