PulseExploreJournal ClubDebatesTrendingResearchersJournals
Instagram
HomeExploreJournal ClubTrending
Synapse
⌘+K
Synapse
February 2, 20260 citations

Machine Learning and Explainable Artificial Intelligence in Carbon Nanotubes Predictions

MOMoacir Guedes OliveiraCACarolina Paula AlmeidaSVSandra M. Venske

Key Points

  • The aim is to improve predictions of carbon nanotube properties through an automated machine learning approach.
  • Utilized automated machine learning techniques to optimize model selection and hyperparameter tuning.
  • Employed multiple machine learning algorithms in an ensemble learning framework.
  • Conducted experiments on CNT datasets with varying target counts and on additional real-world datasets.
  • Implemented an explainable AI method to analyze the predictions.
  • Performed statistical tests to compare the OurAlg algorithm with traditional approaches.
  • OurAlg effectively handled multi-target regression tasks with accurate predictions.
  • Demonstrated favorable performance against classical machine learning algorithms.
  • Provided insights into the explainability of predictions made by the model.
  • Showed robust results across different datasets, indicating its versatility.

Abstract

Nanotechnology is an area that can be applied in all fields of science and nanomaterials, such as Carbon Nanotubes (CNT), have been the subject of studies in order to reduce costs, time for experiments and development. CNTs have diverse applications because they are highly rigid, have high strength, and have low density. Because of this, the evaluation of the properties of carbon nanotubes and identification of their coordinates are important fields of research for science. Many real problems predictions, including those on carbon nanotubes, are Multi-Target Regression (MTR) tasks, where multiple continuous target variables are predicted. Machine learning techniques are well suited to address these types of problem. However, the literature reports approaches composed of complex models, which are manually configured and do not present any level of explainability. The goal of this work is to use an Automated Machine Learning (AutoML) technique, for which 10 classic machine learning algorithms are available. The proposal, named OurAlg, uses ensemble learning and AutoML to choose an ensemble of algorithms and their hyperparameters to handle MTR problems. Three CNT data sets were used in the experiments, with 3 and 4 targets, to predict the coordinates and properties of the carbon nanotubes. We also tested our approach on a set of 18 real-world data sets, with different numbers of targets, to evaluate its performance. In addition, an explainable artificial intelligence method was applied to CNT problems. Statistical tests were used and the OurAlg algorithm was compared favorably with classical algorithms and also with the literature.

Ask AI
Helpful
Bookmark
Share
View Full Paper

Cite This Study

Oliveira et al. (2026) studied this question.

synapsesocial.com/papers/6980ffe7c1c9540dea812b95https://doi.org/10.22456/2175-2745.146905
Ask AI
Helpful
Bookmark
Share
View Full Paper