Green extraction of bioactive compounds from orange peel waste using NADES and microwave-assisted technique: A CatBoost-GMDH ensemble optimized by mantis search algorithm
{"title":"Green extraction of bioactive compounds from orange peel waste using NADES and microwave-assisted technique: A CatBoost-GMDH ensemble optimized by mantis search algorithm","authors":"Mostafa Khajeh , Mansour Ghaffari-Moghaddam , Jamshid Piri , Afsaneh Barkhordar , Didem Saloglu","doi":"10.1016/j.jarmap.2025.100668","DOIUrl":null,"url":null,"abstract":"<div><div>The increasing need for sustainable and efficient extraction methods has led to interest in green technologies for the extraction of bioactive compounds from agricultural waste. Orange peel, a rich of polyphenols and flavonoids, offers significant potential for sustainable application. This research formulated a novel multi-task optimization approach combining the Mantis Search Algorithm (MSA) with hybrid machine learning models to extract bioactive compounds from orange peel waste using natural deep eutectic solvents (NADES) and microwave-assisted extraction (MAE). The holistic strategy employed a multi-task optimization approach that concurrently optimized four key components: feature selection, model hyperparameters, ensemble weights, and process parameters. The process parameters investigated included microwave power (302–495 W), extraction temperature (31–59 °C), extraction time (5.2–30 min), and mass-to-solvent ratio (41–80 mg/mL). Three machine learning models were developed and systematically compared: CatBoost, Group Method of Data Handling (GMDH), and their weighted ensemble fusion. The ensemble MSA-hybrid model exhibited the best predictive performance with R² of 0.656, 0.981, and 0.990 for total phenolic content, total flavonoid content, and DPPH radical scavenging activity, respectively. Temperature was found to be the most significant process parameter for all response variables, followed by extraction time and mass-to-solvent ratio. The multi-task optimization approach successfully developed robust predictive models capable of guiding extraction parameter selection for improved bioactive compound yields. Extensive validation using thorough residual analysis, stability testing, and confidence interval analysis reaffirmed model reliability and generalizability. This novel study was successful in offering valuable industry-ready solutions for sustainable bioactive compound extraction while supporting agricultural waste valorization and circular economy concepts.</div></div>","PeriodicalId":15136,"journal":{"name":"Journal of Applied Research on Medicinal and Aromatic Plants","volume":"49 ","pages":"Article 100668"},"PeriodicalIF":3.6000,"publicationDate":"2025-10-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Journal of Applied Research on Medicinal and Aromatic Plants","FirstCategoryId":"97","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S2214786125000488","RegionNum":2,"RegionCategory":"农林科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"PLANT SCIENCES","Score":null,"Total":0}
引用次数: 0
Abstract
The increasing need for sustainable and efficient extraction methods has led to interest in green technologies for the extraction of bioactive compounds from agricultural waste. Orange peel, a rich of polyphenols and flavonoids, offers significant potential for sustainable application. This research formulated a novel multi-task optimization approach combining the Mantis Search Algorithm (MSA) with hybrid machine learning models to extract bioactive compounds from orange peel waste using natural deep eutectic solvents (NADES) and microwave-assisted extraction (MAE). The holistic strategy employed a multi-task optimization approach that concurrently optimized four key components: feature selection, model hyperparameters, ensemble weights, and process parameters. The process parameters investigated included microwave power (302–495 W), extraction temperature (31–59 °C), extraction time (5.2–30 min), and mass-to-solvent ratio (41–80 mg/mL). Three machine learning models were developed and systematically compared: CatBoost, Group Method of Data Handling (GMDH), and their weighted ensemble fusion. The ensemble MSA-hybrid model exhibited the best predictive performance with R² of 0.656, 0.981, and 0.990 for total phenolic content, total flavonoid content, and DPPH radical scavenging activity, respectively. Temperature was found to be the most significant process parameter for all response variables, followed by extraction time and mass-to-solvent ratio. The multi-task optimization approach successfully developed robust predictive models capable of guiding extraction parameter selection for improved bioactive compound yields. Extensive validation using thorough residual analysis, stability testing, and confidence interval analysis reaffirmed model reliability and generalizability. This novel study was successful in offering valuable industry-ready solutions for sustainable bioactive compound extraction while supporting agricultural waste valorization and circular economy concepts.
期刊介绍:
JARMAP is a peer reviewed and multidisciplinary communication platform, covering all aspects of the raw material supply chain of medicinal and aromatic plants. JARMAP aims to improve production of tailor made commodities by addressing the various requirements of manufacturers of herbal medicines, herbal teas, seasoning herbs, food and feed supplements and cosmetics. JARMAP covers research on genetic resources, breeding, wild-collection, domestication, propagation, cultivation, phytopathology and plant protection, mechanization, conservation, processing, quality assurance, analytics and economics. JARMAP publishes reviews, original research articles and short communications related to research.