Nowadays, microfinance companies offer more lending to a wider range of people. Some ethical microfinance lending requires accurate risk models based on the borrower’s financial history. Many traditional credit scoring methods fail to accurately identify the financial conditions of fraudulent users. The MicroFin-CNet, proposed in this paper, introduced a new Fuzzy Neural Network (FNN) that estimated the true financial situation of the borrower by capturing the complex socio-economic patterns of low-income and unbanked populations, thereby reducing biased or ineffective lending decision-making. The proposed approach is the integrated approach that combines demographic, behavioral, and transactional features through a structured feature embedding layer, followed by multi-scale convolutional learning to capture both local and global risk patterns. The performance of the proposed approach improved by adopting bias-aware normalization and explainable risk indicators that support responsible lending decisions. The classification results show that the proposed approach reduces the false rejections of low-risk borrowers. The results show that the proposed approach, MicroFin-CNet, achieved high classification performance.