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。国内普惠金融概念的引入始于2006年的亚洲小额信贷论坛会上由中国人民银行研究局副局长焦瑾璞提出了建立普惠制金融体系的概念。普惠金融概念的引入是为了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。本研究主要针对中小企业这一普惠金融的重要客户群体该类企业通常在资金和资源上都略显薄弱但他们在社会经济发展中起着重要作用。特

Inclusive finance is a concept that has developed based on microcredit and microfinance. The introduction of the concept of inclusive finance in China began in 2006 at the Asian "Microcredit Forum," where Jiao Jinping, Deputy Director of the Research Bureau of the People's Bank of China, proposed the concept of establishing an "inclusive" financial system. The introduction of the concept of inclusive finance aims to address the issues of imbalanced and insufficient financial services that are prevalent in China during its rapid economic development. This study focuses on small and medium-sized enterprises (SMEs), an important customer group in inclusive finance. These enterprises often face challenges in terms of funding and resources but play a significant role in socio-economic development. Specifically, all the data used in this study were collected from small and medium-sized enterprises in Xinjiang, which have unique characteristics in terms of regional factors, cultural differences, policy environment, and industrial structure under the background of inclusive finance.

This study first proposes an optimized default risk assessment index system for credit decision-making of small and medium-sized enterprises in Xinjiang. The study collected financial indicator data of selected SMEs from a bank in Xinjiang in 2021, conducted basic descriptive statistics and missing value imputation, and selected the strongest individual indicator for default identification and the optimal combination of indicators for overall default identification through indicator predictive ability analysis, multicollinearity diagnosis, and feature selection. This helps further discriminate the default status of enterprises and evaluate their credit risk. For the actual dataset, the established optimal credit indicator system achieves a classification accuracy rate of 95.85% with the assistance of a simple machine learning model, demonstrating its great potential in credit rating research and application in Xinjiang.

Secondly, considering the optimization and interpretability of the indicator system, this study constructs suitable default prediction and credit rating models as the second part of the research. When constructing the default prediction model, we propose a novel CT-XGBoost prediction model that combines cost sensitivity and threshold methods based on the supervised learning concept to address the issue of class imbalance in the credit default dataset. This model improves XGBoost through these two algorithmic strategies and solves the challenges of correctly allocating the misclassification costs of the two classes and setting reasonable thresholds. The experiment uses a loan default database of a bank in Xinjiang from 2017 to 2021. The CT-XGBoost model established in the experiment has an average AUC value of 96.38%, outperforming other default prediction models (AUC values ranging from 90.35% to 95.44%). The results show that the supervised learning model established in this study, with the two algorithmic strategies of cost sensitivity and threshold methods, outperforms existing general models in dealing with class imbalance problems.

When constructing the credit rating model, we adopt the unsupervised learning concept of anomaly detection and develop a novel credit safety risk warning system. The system consists of three parts: data preprocessing, feature extraction, and construction of enterprise credit risk classification models. The experimental data include the Dkxx dataset generated from the operations of a bank in Xinjiang, as well as publicly available datasets such as LendingClub credit data, German credit data, and Thelrish Dummy Bank credit data. In the data preprocessing stage, financial data is texturized and an information matrix is constructed. The weighted system is used to remove irrelevant information and undetected indicators and redundant symbols. To address the issue of imbalanced financial credit data samples, we use the following approach: after feature ranking using XGBoost, we use the SMOTE method to augment normal samples, reducing the class imbalance problem to an advantage in anomaly detection. In the feature extraction stage, denoising autoencoders (DAE) are used to classify enterprises based on reconstruction errors, improving the model's robustness. The credit risk assessment level is determined using a multi-level credit safety risk classification based on comprehensive indicators. Finally, warning threshold values are introduced for warning analysis, and the prediction results generated by the constructed model are adjusted by an expert group to ensure the accuracy and reliability of the system.

Furthermore, to further improve discrimination accuracy and expand data sources, this study introduces credit investigation reports as unstructured textual data for credit default prediction and explores their role in improving prediction effectiveness. The research data used includes enterprise credit information from a rural credit cooperative in Xinjiang, China. Text data analysis is conducted from two aspects: text attributes and text topics. Relevant information is extracted, and models such as logistic regression, support vector machines, and neural networks are used for prediction. The results show that the textual information in credit investigation reports provides incremental information for credit default prediction based on quantitative financial indicators. Considering both text attribute indicators and text topic indicators alongside financial indicators achieves the best credit default prediction performance.

In conclusion, this study provides valuable practical experience for credit decision-making for SMEs, improves the interpretability of the model by combining actual indicator systems, and provides more accurate and reliable credit decision models for potential future applications in financial institutions. Additionally, this study presents prospects for the field of credit evaluation, such as strengthening research on textual data and optimizing rating indicator systems using multiple data sources. This study has made beneficial explorations in the credit evaluation of SMEs, contributing actively to the development of inclusive finance and the improvement of credit evaluation accuracy and reliability.