I. INTRODUCTION
In the stage where credit-based economic is developing rapidly, the borrowing and lending operations of enterprises have a significant pull on the economy. Among them, micro, small and medium-sized enterprises (MSMEs), despite their usually small business scale, are more active and play an important part of all types of economies [1], [2]. Considering the credit risk, banks generally provide different products and services for different volume of customers, which is consistent with the credit rationing theory [3]. Due to the general lack of collateral, small scale of operation, irregular corporate financial system, and low information transparency of MSMEs, banks provide them with more limited products and services [4]. Because the objective information asymmetry in the market economy, it is difficult for MSMEs to obtain sufficient financial support and for banks to find quality customers and expand their credit business. [5], [6] Currently, artificial intelligence is developing rapidly, and artificial neural network algorithms are a key component of research in the field of artificial intelligence [7]. The combination of artificial neural networks and bank credit rationing is an effective way to solve the traditional credit decision problem. The purpose of this research is to establish a credit risk assessment model for MSMEs based on the principle of BP neural network learning algorithm by studying the data mining of unprocessed information about corporate credit, and to establish a credit decision model by iteratively solving a nonlinear programming problem in a real lending market context using the interior point method. The artificial neural network has good adaptive learning capability. It works mainly by simulating biological neural networks, but the drawback is its poor ability to express special contexts [8]. Therefore, considering that the credit market is permanently affected by unexpected factors in all aspects of the economy and society, the credit strategy obtained based on the decision model is adjusted practically to obtain a fairer decision result according to the firm's own characteristics.