Scholarlink

<p>The convergence of artificial intelligence (AI), machine learning (ML), and environmental, social, and governance (ESG) considerations has transformed financial decision-making. This transformation yields several advantages, including enhanced predictive accuracy, real-time fraud detection, expanded access for underserved populations, integration of sustainability metrics into credit models, and increased transparency and regulatory compliance. This systematic review addresses four research questions using the Antecedents-Decisions-Outcomes (ADO) framework to map, synthesize, and critically evaluate the AI/ML-ESG nexus within marketplace lending. Following PRISMA-2020 guidelines, 555 peer-reviewed studies published between January 2015 and December 2025 were identified from Scopus and Web of Science and analyzed through narrative synthesis. Methodological trends have shifted from statistical approaches (65% through 2017) to machine learning (2018–2021), deep learning (2020–2023), and, most recently, explainable AI with ESG integration (42% of studies published from 2024 onward). Based on descriptive comparison of individually reported results across heterogeneous studies, ensemble machine learning methods demonstrate superior performance (88–96% accuracy, AUC 0.88–0.96) and efficiency (0.5–4&#xa0;h, approximately $8 per application) compared to traditional statistics (65–75% accuracy, 8–48&#xa0;h, $45–125). Deep learning techniques also achieve high accuracy (85–95%), while the limited number of ESG-ML hybrid models report balanced sustainability objectives with robust performance (87–93%). However, ESG rating inconsistency across agencies (correlation 0.38–0.59) poses a critical challenge to the reliability of ESG-ML hybrid models, and the substantial heterogeneity across datasets, evaluation protocols, and geographic contexts limits the generalizability of cross-study performance comparisons. Six critical research gaps are identified: explainability, ESG standardization, real-time model adaptability, fairness, privacy, and alternative data validation. The findings indicate that the field is advancing toward responsible, transparent lending practices, though significant methodological and standardization challenges remain.</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

A systematic review of artificial intelligence, machine learning, and environment–social–governance in marketplace lending

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月18日

查看全文
  • Jewel Kumar Roy

摘要

The convergence of artificial intelligence (AI), machine learning (ML), and environmental, social, and governance (ESG) considerations has transformed financial decision-making. This transformation yields several advantages, including enhanced predictive accuracy, real-time fraud detection, expanded access for underserved populations, integration of sustainability metrics into credit models, and increased transparency and regulatory compliance. This systematic review addresses four research questions using the Antecedents-Decisions-Outcomes (ADO) framework to map, synthesize, and critically evaluate the AI/ML-ESG nexus within marketplace lending. Following PRISMA-2020 guidelines, 555 peer-reviewed studies published between January 2015 and December 2025 were identified from Scopus and Web of Science and analyzed through narrative synthesis. Methodological trends have shifted from statistical approaches (65% through 2017) to machine learning (2018–2021), deep learning (2020–2023), and, most recently, explainable AI with ESG integration (42% of studies published from 2024 onward). Based on descriptive comparison of individually reported results across heterogeneous studies, ensemble machine learning methods demonstrate superior performance (88–96% accuracy, AUC 0.88–0.96) and efficiency (0.5–4 h, approximately $8 per application) compared to traditional statistics (65–75% accuracy, 8–48 h, $45–125). Deep learning techniques also achieve high accuracy (85–95%), while the limited number of ESG-ML hybrid models report balanced sustainability objectives with robust performance (87–93%). However, ESG rating inconsistency across agencies (correlation 0.38–0.59) poses a critical challenge to the reliability of ESG-ML hybrid models, and the substantial heterogeneity across datasets, evaluation protocols, and geographic contexts limits the generalizability of cross-study performance comparisons. Six critical research gaps are identified: explainability, ESG standardization, real-time model adaptability, fairness, privacy, and alternative data validation. The findings indicate that the field is advancing toward responsible, transparent lending practices, though significant methodological and standardization challenges remain.