Scholarlink

<p>An efficient and affordable e-nose system is crucial for advancing artificial olfaction across domains. This study proposes a low-cost e-nose system featuring seamless LabVIEW-based data acquisition, Topological Data Analysis (TDA)-powered feature extraction, and Machine Learning (ML)-based classification to classify four Volatile Organic Compounds (VOCs): acetone, toluene, isopropanol, and ethanol. TDA, combined with hand-crafted features and advanced tools, identifies key features to create unique VOC fingerprints for successful classification by ML algorithms. The random forest algorithm achieved 89% accuracy on the raw data via fivefold cross-validation for VOC classification. On the other hand, on noisy data, the standalone random forest achieved an accuracy of 41.83%, while the TDA-augmented random forest improved it to 92.31%. TDA also outperformed Principle Component Analysis (PCA) in denoising, with TDA–ML achieving 91.67% accuracy compared to 41.85% with PCA–ML. Thus, our framework with TDA and ML promises to be a better solution for a noisy low-cost e-nose system.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

The synergy of topological data analysis and machine learning for low-cost e-nose systems

  • 期刊论文

  • 发表日期：

    2025年3月6日

查看全文
  • R. Shylaja,
  • D. Nedumaran,
  • C. Venkateswaran

摘要

An efficient and affordable e-nose system is crucial for advancing artificial olfaction across domains. This study proposes a low-cost e-nose system featuring seamless LabVIEW-based data acquisition, Topological Data Analysis (TDA)-powered feature extraction, and Machine Learning (ML)-based classification to classify four Volatile Organic Compounds (VOCs): acetone, toluene, isopropanol, and ethanol. TDA, combined with hand-crafted features and advanced tools, identifies key features to create unique VOC fingerprints for successful classification by ML algorithms. The random forest algorithm achieved 89% accuracy on the raw data via fivefold cross-validation for VOC classification. On the other hand, on noisy data, the standalone random forest achieved an accuracy of 41.83%, while the TDA-augmented random forest improved it to 92.31%. TDA also outperformed Principle Component Analysis (PCA) in denoising, with TDA–ML achieving 91.67% accuracy compared to 41.85% with PCA–ML. Thus, our framework with TDA and ML promises to be a better solution for a noisy low-cost e-nose system.