A new systematic review published in the journal AI in Civil Engineering outlines a strategic roadmap for integrating Explainable Artificial Intelligence (XAI) into housing manufacturing to reduce carbon dioxide emissions. The study, authored by Miguel Mora and Pingbo Tang, addresses a critical gap in the construction industry: while AI algorithms can optimize production schedules to lower energy use, their opaque “black box” nature prevents human operators from trusting or verifying the results. By applying the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) methodology, the researchers synthesized existing literature to define how XAI can make these AI-driven decisions transparent, thereby enabling the decarbonization of housing production.
The construction industry faces significant environmental challenges, with emissions from the fabrication of construction materials accounting for 37% of global CO2 emissions according to the 2022 Global Status Report for Buildings and Construction. In manufacturing contexts, researchers have long identified that optimizing machinery usage through production scheduling can reduce these emissions. However, the housing manufacturing sector presents unique complexities due to high customization requirements and uncertainties arising from human, machine, and business processes. The review posits that integrating manufacturing principles with AI offers a promising pathway to mitigate these emissions, provided that the AI systems are explainable to the users who remain ultimately responsible for production outcomes.
The core technical challenge lies in the Job-Shop Scheduling Problem (JSSP), an NP-hard optimization task that requires extensive computational effort. In housing manufacturing, this is further complicated by the need for flexible production lines, leading to the Flexible Job-Shop Scheduling Problem. Traditional AI approaches, such as Genetic Algorithms and Artificial Neural Networks, can propose solutions in reasonable timeframes but operate based on trained data rather than physical phenomena. This creates a transparency problem: the AI defines weights and relationships internally, withholding the reasoning behind its optimization choices from the human user who must account for the final production schedule.
Explainable Artificial Intelligence aims to resolve this transparency deficit by providing clear and understandable explanations for AI actions and decisions. The review classifies XAI approaches based on three primary dimensions: stage, scope, and output. The stage refers to whether the explanation is generated ante-hoc by the AI model itself or post-hoc by an external model. The scope indicates whether the explanation is global, providing a user-interpretable representation of the model’s overall reasoning, or local, focusing on the reasoning behind a specific result. The output describes how the explanation is presented, which can be numeric, textual, graphical, or a mix of these formats.
The researchers conducted an initial search query in the Google Scholar database that returned 30,700 results. After sorting by relevance, the team analyzed the first 210 papers and reports. A subsequent, more targeted search focused on manufacturing and construction contexts, yielding 37 papers. From this set, the researchers selected 14 papers for detailed analysis based on their relevance to manufacturing processes or the construction industry. Of these, eight were literature reviews of XAI, while six focused specifically on XAI to support decision-making in manufacturing contexts. This rigorous filtering process ensured that the synthesis was grounded in the most applicable state-of-the-art methodologies.
Analysis of the six selected manufacturing-focused studies revealed significant gaps in the current XAI landscape. Only one of the six studies tested and validated its XAI approach with an expert panel, highlighting a broader lack of user-centric validation in the field. Furthermore, only one study specifically addressed XAI in the context of production scheduling, and just one applied XAI to explain the relationship between sustainability features and operational variables in manufacturing. These findings underscore that while XAI is prevalent in industries like healthcare and banking, its application to the specific problem of CO2 reduction in housing manufacturing remains largely unexplored.
Despite these gaps, the review identified several actionable recommendations for developing effective XAI systems. Three of the six studies emphasized the importance of feature-importance ranking, suggesting that identifying which variables most significantly impact the AI’s decisions enhances transparency. Four of the six studies applied the Shapley Additive Explanations (SHAP) methodology, a model-agnostic technique that uses coalitional game theory to rank feature importance. Additionally, three studies developed explanation interfaces using text, images, or charts, indicating that graphical interfaces are crucial for supporting the presentation of AI reasoning, assumptions, and results to users.
The review also highlighted the necessity of a post-hoc approach to XAI. All six selected studies applied post-hoc explanations, where an external model generates the explanation rather than the AI model itself. This is recommended because the primary AI model is designed to solve the optimization problem, not to support user understanding.
Implementing XAI in housing manufacturing faces several practical challenges, primarily related to data availability. XAI approaches require training data, such as real-time measurements of machine energy consumption, to characterize CO2 emissions. However, data specifically regarding CO2 emissions from machinery in housing manufacturing is limited. The researchers suggest that collecting energy data from machinery brochures and on-site measurements can help estimate CO2 emissions. On-site monitoring provides more accurate results, enabling the characterization of task-level energy consumption and the subsequent calculation of associated carbon footprints using established emission factors.
Another critical challenge is the integration of contextual data specific to the housing manufacturing industry. The XAI approach must reflect the specific constraints of the problem, such as the size and weight of fabricated panels, which constrain inventory, and the continuous-flow mode of production lines. These restrictions structure the AI scheduling approach and must be integrated into the optimization process. Without incorporating these industry-specific variables, users may not validate the XAI as a proper representation of factory-floor reality, leading to a failure in adoption. The review emphasizes that a one-size-fits-all XAI approach is insufficient; instead, solutions must be designed explicitly for domain-specific problems.
The study proposes a workflow and roadmap for defining an XAI to support housing manufacturing decarbonization. This roadmap begins with identifying industry-specific requirements in housing manufacturing, followed by selecting and understanding the most appropriate AI methodology. Finally, it involves defining the most appropriate XAI to support the decarbonization goals. This process requires a socio-technical perspective, integrating human considerations into the technological implementation. Understanding the multi-level decisions made on the factory floor, from operators to managers, is essential for defining user-centric requirements that align with the scope of each user’s decision-making authority.
In conclusion, the review demonstrates that while AI can support decarbonization in production scheduling, the lack of explainability hinders its practical application in housing manufacturing. By adopting a user-centric perspective and integrating industry-specific knowledge, researchers can develop XAI approaches that make AI reasoning transparent. This transparency is crucial for enabling the integration of AI-assisted decarbonization approaches in the production scheduling of housing manufacturing. The findings of this study, while focused on housing, offer insights applicable to the development of domain-specific XAI approaches in other industries seeking to balance operational efficiency with environmental sustainability.
Subject of Research: Explainable AI for CO2 emissions reduction in housing manufacturing: a review
Article Title: Explainable AI for CO2 emissions reduction in housing manufacturing: a review
Article References: Mora, M., & Tang, P. (2026). Explainable AI for CO2 emissions reduction in housing manufacturing: a review. AI in Civil Engineering, 5(1), Article 4. https://doi.org/10.1007/s43503-026-00086-w
Image Credits: AI Generated
DOI: 10.1007/s43503-026-00086-w
Keywords: Explainable, emissions, reduction, housing, manufacturing, review, scientific research
News Source: Denise Maddox. (November 8, 2026). Review Maps Path for Explainable AI in Green Housing. Scienmag.
