Scholarlink

<p>Street vendors in India’s informal economy remain largely excluded from digital finance due to severe data scarcity, undocumented income streams, and minimal participation in formal financial systems. This exclusion is compounded by low digital literacy, sporadic earnings, and persistent reliance on cash transactions—particularly in Odisha, where the informal sector dominates urban employment. As a result, financial institutions struggle to extend inclusive services such as credit and insurance to these underserved communities. To address this challenge, we propose a generative AI framework based on conditional generative adversarial networks (cGANs) to simulate realistic financial behaviors from limited vendor data. Trained on transaction data from 1000 urban vendors, our model generates synthetic profiles conditioned on socio-demographic attributes, achieving high fidelity (Fréchet distance = 10.12) and behavioral diversity (entropy = 3.59 bits). Augmenting real-world datasets with these synthetic records improves loan-default prediction by 3.2% and anomaly detection by 2.9%. The framework incorporates fairness-aware training, maintains demographic parity (deviation &lt; 2.4%), and has been validated through stakeholder engagement. These results show that responsibly applied generative AI can enhance data equity and foster inclusive digital finance in low-resource settings.</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

Conditional GAN-based synthetic financial modeling for street vendors in India’s informal economy

  • 期刊论文

  • 发表日期：

    2025年6月23日

查看全文
  • Swachha Sisir Das,
  • Sasmita Mishra,
  • Zefree Lazarus Mayaluri

摘要

Street vendors in India’s informal economy remain largely excluded from digital finance due to severe data scarcity, undocumented income streams, and minimal participation in formal financial systems. This exclusion is compounded by low digital literacy, sporadic earnings, and persistent reliance on cash transactions—particularly in Odisha, where the informal sector dominates urban employment. As a result, financial institutions struggle to extend inclusive services such as credit and insurance to these underserved communities. To address this challenge, we propose a generative AI framework based on conditional generative adversarial networks (cGANs) to simulate realistic financial behaviors from limited vendor data. Trained on transaction data from 1000 urban vendors, our model generates synthetic profiles conditioned on socio-demographic attributes, achieving high fidelity (Fréchet distance = 10.12) and behavioral diversity (entropy = 3.59 bits). Augmenting real-world datasets with these synthetic records improves loan-default prediction by 3.2% and anomaly detection by 2.9%. The framework incorporates fairness-aware training, maintains demographic parity (deviation < 2.4%), and has been validated through stakeholder engagement. These results show that responsibly applied generative AI can enhance data equity and foster inclusive digital finance in low-resource settings.