普惠金融是在小额信贷和微型金融基础上发展出的一种金融体系概念致力于解决中国在经济高速发展的过程中普遍存在的金融服务不平衡、不充分等问题。由于中小企业通常在资金、资源上略显薄弱的同时在社会经济发展中起着重要作用本研究主要针对中小企业这一普惠金融的重要客户群体探索金融机构的信贷决策方法。这种方法应适应市场需求和技术变革的同时不能牺牲金融机构的盈利能力和风险管理。本文选取新疆地区的中小型企业数据进行实证

Inclusive finance is a concept of financial system that has developed based on small-scale credit and microfinance, aiming to address the issues of imbalanced and insufficient financial services in China during its rapid economic development. Small and medium-sized enterprises (SMEs) usually play an important role in the social and economic development while facing challenges in terms of funding and resources. This study focuses on SMEs as an important customer group of inclusive finance and explores the credit decision-making methods of financial institutions. These methods should adapt to market demands and technological changes without sacrificing the profitability and risk management of financial institutions. This article selects data from small and medium-sized enterprises in Xinjiang region for empirical research, and the main research work and innovations can be summarized in the following four aspects:

(1) This article proposes a dynamic and optimized default risk assessment index system to address the insufficient ability of existing credit assessment index systems to distinguish significantly between default and non-default SMEs, and the problem of redundancy and overlap between rating indicators. The research is based on dynamic feature selection using multiple methods to identify the combination of single default indicators with the strongest discrimination ability and the overall default identification ability, in order to establish a credit rating feature set with optimal default discrimination ability. Based on the financial indicators data of SMEs from a bank in Xinjiang in 2021, empirical research found that the optimal feature set achieved an accuracy rate of 95.85% and an AUC value of 0.8504 for the test set, demonstrating good model robustness. This method provides a feature set method system for credit risk evaluation of SMEs that can be applied by banks and other financial institutions, and provides ideas for further improving model interpretability.

(2) This article proposes a novel and efficient prediction model called CT-XGBoost to address the issue of class imbalance in credit default datasets and the resulting model bias. This model is an improvement of the strong classification model XGBoost, using cost-sensitive strategies and threshold methods. The study selected five commonly used credit default prediction models as benchmark models, including logistic regression, SVM, neural network, random forest, and XGBoost, and studied the impact of parameter settings on the performance of the improved model through sensitivity analysis. The experiment used a loan default database from a bank in Xinjiang from 2017 to 2021. The average AUC value of the CT-XGBoost model established in the experiment was 96.38%, which was better than other default prediction models (AUC value ranging from 90.35% to 95.44%). The results demonstrate that this research can effectively estimate the credit risk of SMEs, provide suggestions for SMEs to fully utilize loan funds to avoid credit risks, help credit rating agencies improve their credit rating systems by providing interpretable reasons to ensure system reliability, and effectively solve the issue of class imbalance in data, providing new ideas for credit risk assessment of SMEs.

(3) In order to further address the issue of class imbalance and considering the context of inclusive finance, this study aims to minimize costs, including the high costs of manual calculation and training label input in the initial modeling process. It proposes a credit rating model for SMEs based on anomaly detection, which can be classified as an unsupervised learning method less used in the field of credit assessment. The system content is mainly divided into three parts: data preprocessing, feature extraction, and construction of enterprise credit risk classification models. After feature ranking, the normal samples are augmented using the Smote method, and the feature extraction stage uses denoising autoencoders (DAE) to classify enterprises based on reconstruction errors, thereby improving the robustness of the model. The model uses a multi-level credit safety risk level division based on comprehensive indicators, and introduces warning thresholds for warning analysis. The prediction results generated by the model are corrected by an expert group to ensure the accuracy and reliability of the system. The results demonstrate the excellent application prospects of anomaly detection methods in the field of credit decision-making, and provide effective empirical evidence for the application of unsupervised learning in the field of credit decision-making for SMEs.

(4) In the context of inclusive finance, missing user information is a common practical problem. In order to improve discrimination accuracy and expand data sources, this study introduces credit investigation reports as unstructured textual data for credit default prediction, exploring their role in improving prediction effectiveness. The research data used enterprise credit information from a rural credit cooperative in Xinjiang, China. The textual data is analyzed from two aspects: textual attributes and textual themes, to extract relevant information. Models such as logistic regression, support vector machine, and neural network are used for prediction. The experimental results show that by adding textual attribute indicators and textual theme indicators to the financial indicators, the optimal AUC reaches 80.16%, a 13.75% improvement over the baseline model. The results indicate that the textual information in credit investigation reports can provide incremental information for credit default prediction based on quantitative financial indicators, and considering both textual attribute indicators and textual theme indicators on top of financial indicators can achieve the best credit default prediction performance.

In conclusion, this study provides valuable practical experience for credit decision-making for SMEs. It improves the interpretability of models based on actual indicator systems and provides more accurate and reliable credit decision-making models for potential practical applications in financial institutions in the future. At the same time, it also provides prospects for the future development of credit evaluation field, makes beneficial explorations in the field of credit decision-making for SMEs, and actively contributes to the promotion of inclusive finance development and the improvement of credit evaluation accuracy and reliability