Authors
Xuhao Guo, Shilin Wen, Zhenjiang Yin, Jianbo Zheng, Jingang Huang, Xianting Lu, Shanshan Qiu, Wei Han, Rongbing Zhou, Xiaobin Xu
Published in
Bioresource technology. Pages 136243. Nov 14, 2026. Epub Nov 14, 2026.
Abstract
Pyrolyzed biochar is widely used in environmental applications, yet its efficacy depends on physicochemical properties that are often unavailable in engineering practice. This necessitates rapid and cost-effective prediction of biochar functional classes from feedstocks and production conditions. This study developed a machine learning framework integrating unsupervised clustering and supervised classification for this purpose. A literature-sourced dataset encompassing biochar properties, feedstocks, modification methods, and pyrolysis conditions was compiled and preprocessed using imputation and Box-Cox transformation. Among the five evaluated unsupervised clustering algorithms, agglomerative clustering (AC), balanced iterative reducing and clustering using hierarchies (BIRCH), and K-Means initially and consistently identified two natural clusters (silhouette coefficient = 0.78, Davies-Bouldin index = 0.43-0.44). The larger cluster was further refined through secondary AC/BIRCH clustering into two subclusters, which together with the smaller cluster yielded three final clusters (Class 0, Class 1, Class 2). These cluster labels were then used as prediction targets for supervised classification with feedstocks, modification methods, and pyrolysis conditions as input features. Among the five supervised classification models, eXtreme gradient boosting (XGBoost) achieved the optimal classification performance, as evidenced by its superior per-class metrics and the highest overall accuracy (87%), with feedstock type and pyrolysis temperature as the most influential features. Independent validation using four biochar samples produced from different feedstocks and pyrolysis conditions showed that the XGBoost predictions were fully consistent with the clustering labels derived from characterization results. This study provides a data-driven approach for predicting biochar functional classes from feedstocks and production conditions, thereby enhancing engineering applicability.
PMID:
43033993
Bibliographic data and abstract were imported from PubMed on 15 Nov 2026.
Read full publication at:
Please sign in
to see all details.
Advertisement
Stats
- Recommendations n/a n/a positive of 0 vote(s)
- Views 5
- Comments 0