Towards net-zero modular construction: machine learning and time series analysis for waste generation and emissions prediction modelling

Purpose – This study aims to address critical sustainability challenges in modular construction by developing a predictive machine learning (ML) framework to estimate off-cut material waste and associated carbon costs during the construction phase. The objective is to enable environmentally and economically informed planning through a decision-support system that integrates ML and time-series forecasting, supporting engineers and contractors in making early-stage decisions aligned with sustainability goals. Design/methodology/approach – A dual-model approach is used. ML algorithms are trained in architectural and material features to predict cut-off waste from modular wall panels, with performance validated using fivefold cross-validation to ensure robustness and generalizability. Concurrently, autoregressive integrated moving average (ARIMA) time-series modeling is applied to forecast greenhouse gas emissions, incorporating a shadow carbon pricing mechanism consistent with Net Zero strategies. The data set comprises 170 modular projects used for model training and validation. Model performance is assessed using mean absolute error, root mean square error, coefficient of determination (R²) and mean absolute percentage error (MAPE). Findings – Gradient boosting achieved the highest predictive accuracy, with material type and room count identified as the key predictors of waste. The model also demonstrated consistent performance across folds with strong accuracy for dominant targets. The ARIMA model effectively projected carbon emissions, achieving a strong MAPE score. The integrated framework successfully links material waste and emissions to carbon costs, enabling early-stage sustainability interventions. The resulting decision-support tool operationalizes these insights, allowing practitioners to predict waste, emissions and costs before execution, thereby enhancing material efficiency, cost planning and regulatory compliance. Research limitations/implications – This study is limited to modular office buildings in the United Arab Emirates, specifically wall panel installations. Broader applicability to other building types and structural components remains to be explored. Future research should focus on generalizing the framework across diverse contexts and integrating adaptive learning models trained on projects from different regions. Practical implications – The proposed framework provides construction professionals with actionable insights for improving material efficiency, forecasting emissions and identifying carbon-related expenditures at the project inception phase. This supports compliance with emerging carbon pricing regulations while improving procurement planning, sustainability integration and waste reduction strategies. Originality/value – This study presents a novel innovative integration of ML and time-series forecasting to simultaneously estimate material waste, carbon emissions and associated economic costs in modular construction. By embedding predictive analytics within a sustainability framework, it provides a scalable tool for minimizing construction waste and internalizing the environmental costs in modular construction planning.