Semisupervised machine learning classification framework for material intensity parameters of residential buildings. (1st August 2021)

• Record Type: Journal Article
• Authors: Vilaysouk, Xaysackda; Saypadith, Savath; Hashimoto, Seiji
• Other Names: Majeau‐Bettez Guillaume guestEditor.; Frayret Jean‐Marc guestEditor.; Ramaswami Anu guestEditor.; Li Yang guestEditor.; Heeren Niko guestEditor.
• Is Part Of: Journal of industrial ecology. Volume 26:Number 1(2022)
• Page Start: 72
• Page End: 87
• Publication Date: 2021-08-01
• Subjects: conditional inference tree -- industrial ecology -- k‐means -- material stock model -- residential buildings -- semisupervised machine learning; Industrial ecology -- Periodicals; Product life cycle -- Environmental aspects -- Periodicals; Industrial management -- Environmental aspects -- Periodicals; Écologie industrielle -- Périodiques; 658.56
• DOI: 10.1111/jiec.13174
• Languages: English
• ISSNs: 1088-1980

Abstract

Abstract: The material intensity (MI) parameter plays an important role when determining amounts of material stocks, material inflows, and material outflows in material stock models. Recently, several studies have summarized MI parameter information for buildings from around the globe into a single database. Nevertheless, insufficiencies of building type information have led to difficulties when using MI data. This study used semisupervised machine learning to classify MI. An open database of MI parameters of buildings was used as input data for our semisupervised machine learning model. We used material composition data of MI as feature data fed into our machine learning (ML) model. Attribute information of those data points belongs to clusters obtained from the original database was used as information to discover building characteristics of buildings in each building of those clusters to assign building labels for data points of the original dataset. Experiment results revealed seven building clusters in the studied dataset. The probability density function of MI of three building clusters follows a Weibull distribution. The remaining clusters follow gamma and lognormal distributions. Building type labels inferred from the results are useful as additional attributes to the original dataset, providing a new dataset of MI that can be adapted easily for other studies when country‐specific MI data are not available. A decision tree for finding appropriate MI parameters was also introduced. The classification model accuracy was 92.73%, which was achieved using only 201 data points. The proposed framework presents possibilities for application to other MI studies. This article met the requirements for a Gold‐Gold JIE data openness badge described at http://jie.click/badges .