Scholarlink

<p>The extraction of <i>Moringa oleifera</i> oil (MOO) from its kernel under microwave irradiation and <i>n</i>-hexane as solvent was modeled using statistical (response surface methodology - RSM) and four machine learning (ML) approaches (gradient boosting with decision trees – GBDT, support vector machine – SVM, artificial neural network – ANN, and extreme learning machine – ELM). The D-optimal design method was used to create twenty-two experiments to assess the impact of extraction time, irradiation power, and solid/solvent ratio on the MOO yield. The characteristics of the ANN, ELM, SVM, and GBDT models were coefficient of determination, <i>R</i><sup>2</sup> = 0.9842, 9340, 8478, 0.8008, and mean relative percentage deviation, MRPD = 2.1393, 5.5857, 13.4217, 14.7639%, respectively, compared to the RSM model with <i>R</i><sup>2</sup> = 0.9568 and MRPD = 4.4.7758%. The results demonstrated the superiority of ANN over RSM, whereas RSM outperformed ELM, SVM, and GBDT. Optimal conditions predicted by RSM and ANN, coupled with the particle swarm optimization tool, are as follows: time = 8&#xa0;min, irradiation power = 540&#xa0;W, and solid/solvent ratio = 1:30&#xa0;g/mL, with corresponding MOO yields of 43.60% and 39.08%. These conditions were validated in the laboratory to be 39.02%. The physicochemical properties of the MOO (acid number = 3.03&#xa0;mg KOH/g, density = 874&#xa0;kg/m<sup>3</sup>, and peroxide value = 8.4 meqO<sub>2</sub>/kg oil) suggest that it could be a viable feedstock for various chemical products.</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

Microwave-Supported Extraction of Moringa Oleifera Oil with N-Hexane: Machine Learning Approaches Versus Response Surface Methodology Modeling

  • 期刊论文

  • 发表日期：

    2026年2月2日

查看全文
  • Babajide Ayobamiji Sotunde,
  • Niyi Babatunde Ishola,
  • Aanuoluwapo Priscilla Abiola,
  • Ayomiposi Samuel Moses,
  • Ayooluwa Paul Ibrahim,
  • Eriola Betiku

摘要

The extraction of Moringa oleifera oil (MOO) from its kernel under microwave irradiation and n-hexane as solvent was modeled using statistical (response surface methodology - RSM) and four machine learning (ML) approaches (gradient boosting with decision trees – GBDT, support vector machine – SVM, artificial neural network – ANN, and extreme learning machine – ELM). The D-optimal design method was used to create twenty-two experiments to assess the impact of extraction time, irradiation power, and solid/solvent ratio on the MOO yield. The characteristics of the ANN, ELM, SVM, and GBDT models were coefficient of determination, R2 = 0.9842, 9340, 8478, 0.8008, and mean relative percentage deviation, MRPD = 2.1393, 5.5857, 13.4217, 14.7639%, respectively, compared to the RSM model with R2 = 0.9568 and MRPD = 4.4.7758%. The results demonstrated the superiority of ANN over RSM, whereas RSM outperformed ELM, SVM, and GBDT. Optimal conditions predicted by RSM and ANN, coupled with the particle swarm optimization tool, are as follows: time = 8 min, irradiation power = 540 W, and solid/solvent ratio = 1:30 g/mL, with corresponding MOO yields of 43.60% and 39.08%. These conditions were validated in the laboratory to be 39.02%. The physicochemical properties of the MOO (acid number = 3.03 mg KOH/g, density = 874 kg/m3, and peroxide value = 8.4 meqO2/kg oil) suggest that it could be a viable feedstock for various chemical products.