Scholarlink

<p>The present study addresses the challenge of cross-interference in taste sensor signals caused by shared features when predicting multiple targets. Specifically, it investigates the simultaneous addition of salt and MSG to aged broth using a voltammetric electronic tongue (E-tongue). The Boruta algorithm was employed to identify independent features responsive to salt or MSG when both were simultaneously added to the aged broth. These features were then used to develop predictive models based on the online sequential extreme learning machine. The constructed models exhibited exceptional predictive performance. For predicting the additive amount of salt, the models achieved <i>R</i>-squared (<i>R</i><sup><i>2</i></sup>) values exceeding 0.98, ratio of performance to deviation (<i>RPD</i>) values greater than 8.5, and range error ratio (<i>RER</i>) values surpassing 24.0. In the case of predicting MSG additive amounts, the models maintained <i>R</i><sup><i>2</i></sup> values above 0.96, with <i>RPD</i> values exceeding 5.0 and <i>RER</i> values greater than 15.0. These findings indicate that integrating machine learning-based feature selection, online learning, and multi-feature taste sensors holds significant potential in enhancing the reliability of predicting seasoning additive qualities in complex foods. This approach provides valuable insights for the application of E-tongue technology in intelligent food manufacturing.</p> Graphical Abstract <p></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

Machine Learning-Enhanced Taste Sensing for Dual Additive Quantification: Simultaneous Salt and MSG Prediction in Complex Food Liquid

  • 期刊论文

  • 发表日期：

    2025年7月2日

查看全文
  • Fangkai Han,
  • Dongjing Zhang,
  • Marwan M. A. Rashed,
  • Chunxia Dai,
  • Xiaorui Zhang,
  • Bianling Jiang,
  • Xingtao Zhang,
  • Xingyi Huang

摘要

The present study addresses the challenge of cross-interference in taste sensor signals caused by shared features when predicting multiple targets. Specifically, it investigates the simultaneous addition of salt and MSG to aged broth using a voltammetric electronic tongue (E-tongue). The Boruta algorithm was employed to identify independent features responsive to salt or MSG when both were simultaneously added to the aged broth. These features were then used to develop predictive models based on the online sequential extreme learning machine. The constructed models exhibited exceptional predictive performance. For predicting the additive amount of salt, the models achieved R-squared (R2) values exceeding 0.98, ratio of performance to deviation (RPD) values greater than 8.5, and range error ratio (RER) values surpassing 24.0. In the case of predicting MSG additive amounts, the models maintained R2 values above 0.96, with RPD values exceeding 5.0 and RER values greater than 15.0. These findings indicate that integrating machine learning-based feature selection, online learning, and multi-feature taste sensors holds significant potential in enhancing the reliability of predicting seasoning additive qualities in complex foods. This approach provides valuable insights for the application of E-tongue technology in intelligent food manufacturing.

Graphical Abstract