Research on Supply Chain Financial Risk Prevention Based on Machine Learning

Artificial intelligence (AI) proves decisive in today's rapidly developing society and is a motive force for the evolution of financial technology. As a subdivision of artificial intelligence research, machine learning (ML) algorithm is extensively used in all aspects of the daily operation and development of the supply chain. Using data mining, deductive reasoning, and other characteristics of machine learning algorithms can effectively help decision-makers of enterprises to make more scientific and reasonable decisions by using the existing financial index data. At present, globalization uncertainties such as COVID-19 are intensifying, and supply chain enterprises are facing bankruptcy risk. In the operation process, practical tools are needed to identify and opportunely respond to the threat in the supply chain operation promptly, predict the probability of business failure of enterprises, and take scientific and feasible measures to prevent a financial crisis in good season. Artificial intelligence decision-making technology can help traditional supply chains to transform into intelligent supply chains, realize smart management, and promote supply chain transformation and upgrading. By applying machine learning algorithms, the supply chain can not only identify potential risks in time and adopt scientific and feasible measures to deal with the crisis but also strengthen the connection and cooperation between different enterprises with the advantage of advanced technology to provide overall operation efficiency. On account of this, the paper puts forward an artificial intelligence-based corporate financial-risk-prevention (FRP) model, which includes four stages: data preprocessing, feature selection, feature classification, and parameter adjustment. Firstly, relevant financial index data are collected, and the quality of the selected data is raised through preprocessing; secondly, the chaotic grasshopper optimization algorithm (CGOA) is used to simulate the behavior of grasshoppers in nature to build a mathematical model, and the selected data sets are selected and optimized for features. Then, the support vector machine (SVM) performs classification processing on the quantitative data with reduced features. Empirical risk is calculated using the hinge loss function, and a regular operation is added to optimize the risk structure. Finally, slime mould algorithm (SMA) can optimize the process to improve the efficiency of SVM, making the algorithm more accurate and effective. In this study, Python is used to simulate the function of the corporate business finance risk prevention model. The experimental results show that the CGOA-SVM-SMA algorithm proposed in this paper achieves good results. After calculation, it is found that the prediction and decision-making capabilities are good and better than other comparative models, which can effectively help supply chain enterprises to prevent financial risks.


Introduction
Te operation and development of an enterprise have been a new topic for a long time, but it is inevitable that an enterprise will inevitably encounter fnancial risks during its process. Risk prevention of an enterprise is a hot topic among many scholars. Te 2022 government work report also repeatedly mentions the risk prevention of enterprises in its overall requirements and development orientation for the development of economic and social. Against the backdrop of the normalization of the COVID-19 epidemic, China successively issued a series of credit policies, tax reduction and fee reduction policies, and tax retention policies for preferential farmers and preferential enterprises. Te relatively loose monetary policy provided a stable guarantee for the continuous operation of enterprises. However, due to the widespread COVID-19 and the instability of the global economy, many supply chain enterprises are still struggling and facing the crisis of bankruptcy crisis. Bloomberg data show that in 2021 there were 121 large-scale bankruptcy cases with debts of more than USD 50 million. It can be seen that the turmoil of the world economic situation and the erosion of the COVID-19 epidemic has brought many challenges to the daily operation of enterprises. Supply chain fnancial risks will not only harm the operation status of enterprises but making it difcult for enterprises to move forward in the operation process. Enterprises will be cautious in their investment and fnancing decisions, and enterprises with more signifcant fnancial risks will even face hidden dangers such as bankruptcy liquidation. Te fnancial and fnancial status of supply chain enterprises is a support for the operation, and an excellent economic situation is a premise for the sustainable development of enterprises. Financial risks will seriously damage the fnancial indicators of enterprises. Terefore, it is imperative for the supply chain must take measures to reduce fnancial risks and strengthen fnancial risk prevention.
Afected by global debt and the COVID-19 epidemic, many supply chain enterprises cannot make ends meet and seek fnancial support from social and fnancial institutions. For minor and moderate-sized enterprises at the tail terminal of the supply chain, it is hard for banks to assess the credit risk of enterprises due to their imperfect management system and capital structure when facing the fnancing problem. Although banks will collect the soft data of enterprises on time to deal with the shortage of credit data, due to the lack of transparency, timeliness, and accuracy of credit data, banks will require higher fnancing costs from fnancing enterprises to reduce their own loan risk, which undoubtedly increases the fnancing burden for enterprises facing the fnancing problem and increases the fnancial market risk faced by enterprises. For the core business of the supply chain business, due to their vast operation system and enormous demand for capital, they are facing more signifcant fnancial risks such as operational risk, credit risk, and settlement risk. Banks will be more stringent in assessing the credit risk of enterprises, and the assessment will be more comprehensive and accurate. Terefore, enterprises should strengthen their eforts to guard against potential fnancial risks. Under the background that the business crisis has an essential infuence on the global economy, it is an important research topic to accurately calculate the fnancial risks faced by enterprises and to explore ways to prevent digital fnancial risks in supply chains.
In the process of operation, the supply chain will face a series of risks, such as market competition risk, the credit risk of trade credit, national related policy risk, national legal risk, manager behavior risk, stakeholder information transmission risk, and natural environment risk. All kinds of threats will harm the operation of the enterprise. Te supply chain should avoid these risks as much as possible when making decisions. Supply chain internal managers will be afected by factors such as market uncertainty, debtors' delay in paying debts, changes in their credit ratings afect the amount of fnancing, and changes in exchange rates resulting in poor foreign exchange translation when making operational decisions, which will cause certain risks to the property operation of enterprises. During the operation process of corporate, the digital fnancial risks faced by enterprises will also be exacerbated by the disagreement of the directorate, conficts between the interests of stakeholders and the beneft of the enterprises, agency problems of senior managers, inefcient investment, and other issues. Due to the diversity and variability of fnancial risk types, it is difcult for managers to make scientifc and reasonable decisions based on personal will. Financial risk prevention (FRP) model has essential signifcance for supply chain enterprises to make fnancial decisions. Te model can consider various objectives of supply chain operation, make use of its advantages of data mining and data processing, visualize all kinds of fnancial data, facilitate managers to make decision analysis based on relevant data, and help leaders to make the best decision. Especially in today's signifcant data era, we should collect and sort out masses of data with the help of relevant models and get the utmost out of the preponderance of big data to help supply chain enterprises make reasonable decisions and promote long-term development.
At present, artifcial intelligence (AI) technology has already been extensively used in diverse felds. Using machine learning algorithms can rapidly and availably arrange data and make decisions. Te model constructed by machine learning algorithms can arrange myriad kinds of matter in supply chain enterprises, such as sales forecast, marketing strategy management, customer support, and order management. Te most applied felds are the media feld and the retail feld. Tis paper uses machine learning algorithms to dispose of the fnancial risk, measures the potential risks faced by supply chain enterprises, helps enterprises make reasonable decisions, and improves the decision-making quality of managers. Collecting and collating the data of the quoted company in the supply chain of China in recent three years, preprocessing the collected data to upgrade the quality, efectively refecting the operating conditions of all aspects of the enterprise, helping to apply the model to identify potential fnancial risks faced by enterprises, and improving the supply chain management capability. Te chaotic grasshopper optimization algorithm can draw the economic characteristic from the selected listed companies' fnancial statements, and the support vector machine algorithm is used to classify the data after feature selection. For the sake of raising the precision of data classifcation, the slime mould algorithm is designed to further process and make the indicators more optimized after data classifcation, on purpose to enhance the demonstration of the decisionmaking system. According to the relevant standards, the fnancial risks that an enterprise may face are classifed, and diferent early warning standards are set, which is helpful for managers to make decisions based on various data parameters, timely discover the risks faced by the enterprise, and take corresponding management measures to prevent hazards.
During the operation of the fnancial risk prevention model (FRP), feature selection (FS) is an essential stage in the model, which is mainly to flter out the critical and 2 Computational Intelligence and Neuroscience irregular features needed for decision-making from broadminded data. Te feature extraction module can be segmented into dependency flter type, wrapper type, and embedded type according to the evaluation status. Among them, packaging will face the problems of maximum processing complexity and learner constraints, while the implementing of embedded is more complicated. In contrast, the dependency flter can calculate the feature subset by using the permanent value and then detect the optimal feature from the accessible features. Tis method has signifcant advantages. Secondly, in the data classifcation module, the hackneyed categorize algorithms are support vector machine, K-nearest neighbor, decision tree, etc. Te K-nearest neighbor algorithm needs to traverse all the data when classifying the data, which has poor timeliness, is sensitive to the K value and greatly infuences on the precision of the ultimate policy decision. Te decision tree algorithm is prone to overftting, especially when there are more data, and the probability of the error rate will increase faster. In the data with a substantial degree of association, the algorithm cannot achieve good results. Te support vector machine algorithm can simplify the classifcation process based on the existing training samples, which can have a good improvement on the efciency and quality of classifcation. Tis method is not sensitive to the selection of kernels. It can facilitate the problematic position of multiclassifcation and multidimensional sample space by constructing the association of support vector machine and has certain robustness. Terefore, compared with the K-nearest neighbor and decision tree methods, support vector machine is more intuitive and accurate in the classifying of parameters and has certain advantages. In this model, the slime mould algorithm can further optimize the classifcation results and enhance the efciency and practicality of the consequence. Te fnancial risk prevention model can perform complex operations of checking economic data and related data, involving production accounting of enterprises, process of working capital, strategic development of organizations, and other aspects. Managers can improve their decision-making quality according to the relevant index parameters of the model, better manage the fnancial risks of the company and strengthen risk prevention. When current scholars study the prevention of supply chain fnancial risks, some scholars expound the factors that afect digital fnancial risks from the angle of empirical analysis and put forward the management and countermeasures to prevent risks; some scholars use machine learning algorithms to analyze supply chain fnancial risks, such as neural network algorithm and TOPSIS method; at present, many scholars use a single machine learning algorithm to solve supply chain fnancial risks. In this study, a combined optimization method is proposed. CGOA-SVM-SMA algorithm can be used to optimize the relevant results. Compared with the previous research, it can better enhance the practicality of the pattern. Te fnancial-riskprevention (FRP) model depends on the artifcial intelligence algorithm proposed in the essay and principally contains four stages: data preprocessing, select features, classifcation, and parameter adjustment. First, collect various fnancial data of the enterprise and preprocess them to enhance the weight of the selected data; secondly, chaotic grasshopper optimization algorithm (CGOA) can choose features and optimize the selection process. Ten, support vector machine (SVM) algorithm can classify the data with reduced features, and the classifcation efciency can be optimized by slime mould algorithm (SMA), to augment the accuracy of the fnal decision. Finally, tests are executed on data sets, and adjustments are made continuously to enhance the expression of the model's prediction decisions. Te main innovations of this essay include the following theories: (1) Innovatively put forward a model to help supply chain enterprises guard against fnancial risks, using the CGOA-SVM-SMA algorithm, which mainly includes the feature selection of CGOA, classifcation of SVM, and parameter adjustment of SMA. (2) Te CGOA algorithm is proposed for feature selection and optimization, which can efectively drop the sophistication of data calculation and enhance the prediction function of the pattern. Te algorithm is applied to the novel hinterland of supply chain fnance. (3) Te slime mold algorithm can refner the classifcation process of the pattern to enhance the management efect and decision-making performance of the pattern, and the innovative algorithm is used to refne the classifcation process to improve the efciency. (4) Te algorithm of SMA-SVM is proposed in the classifcation process, in which the SVM is used to classify and sort the data after feature selection. Ten, the SMA is used to optimize these data, to strengthen the usefulness of the model. Te former literature seldom uses SMA to enhance the specifcations of the SVM, and the new model suggested in this study can signifcantly enhance the catalog efciency and accuracy; (5) Te rapid development of fnancial technology makes supply chain digital fnance face unprecedented risks. Te degree of integration between fnancial technology and supply chain fnance in previous studies is not high. Te innovative digital fnancial risk prevention model constructed in this study can help supply chain enterprises to be timely perceive potential risks in the Internet era with the advantages of fnancial technology. Te model has high accuracy in prediction and decision-making and has specifc practical value. (6) Te machine learning algorithm is applied to supply chain fnancial risk management. Tis paper breaks the previous single research method, explores the machine learning algorithm combination optimization model, solves the management problem of the actual operation of the enterprise, and constructs the enterprise risk prevention and early warning mechanism, which helps the enterprise to prevent Computational Intelligence and Neuroscience fnancial risks, enriches the application scope of the machine learning model, and solves practical problems.
Based on the above-given analysis, the overall organization of the essay is as follows. Because of the present situation of introducing the research background of this paper in Section 1; Section 2 introduces the current work related to fnancial risk prevention, as the research basis of this paper, which is helpful for further analysis in the light of the current research in the following part; next, Section 3 constructs the fnancial risk prevention model and elaborates the model in detail; In Section 4, relevant data are selected to verify the proposed model, which proves the feasibility of the model. Finally, Section 5 is a summary and outlook, which summarizes the paper and points out the research localization and the research feld in the future.

Related Work
Economic globalization has increased communication among countries, making capital markets more active. With the increase in the amount of capital in circulation, the capital requirement of enterprises is augmenting increasingly. But in the meantime, it has also expanded the impact of adverse events such as the COVID-19 epidemic, causing turmoil in supply chain fnancial markets. Enterprises are in urgent need of fnancial risk prevention measures to deal with the impact of globalization.

Evaluation of Supply Chain Digital Financial Risk Prevention System.
Using the advantages of AI and ML algorithms, a fnancial risk prevention model is built to study the infuential risk factors afecting the operation of supply chain enterprises, and a judge target setup of enterprise monetary afairs risk is constructed, which can efcaciously heighten the prediction accuracy and decision-making correctness [1]. In the territory of fnance of supply chain, with the development of the digital economy, the application of digital currency is more and more extensive, and its impact on the social economy is also more prominent. Tis emerging digital currency has gradually exposed its risks as it continues to grow [2]. To accurately predict the digital decision-making currency and reduce its operational risk, an improved deep neural network algorithm (DNN) is adopted, which uses information such as transactions and monetary returns to extract the relevant characteristics of Bitcoin to estimate and predict the price of Bitcoin, thus contributing to better e-commerce decision-making and reducing fnancial risk [3].
Machine learning algorithm has now been proverbially utilized in various domains. Applying it to the supply chain fnance feld can help realize the changing and escalating of business intelligence, which meets the requirements of the industry 4.0 era. Financial risk is a signifcant problem faced by many supply chain enterprises in the current period, which also afects social and economic development [4]. Applying the ML algorithm to the fnancial risk management of companies can get the utmost out of the advantages of the machine learning algorithm to solve the decisionmaking problems facing the supply chain. Te machine learning algorithms can manage market risk, credit risk, etc. However, it is currently facing challenges in data, algorithms, and models in the application process. It needs to continuously optimize its algorithm to solve practical problems [5]. Some scholars conducted a systematic review of 232 studies, analyzed the necessity and urgency of applying artifcial intelligence technology to identify the fnancial difculties of supply chains, and pointed out that in data preprocessing, attention should be paid to data balance and dimensionality reduction, and the evolution index should be optimized to improve the model performance [6]. Some scholars also apply the machine learning algorithm to the operational risk management faced by the enterprises in the operation course. In manufacturing enterprises, the machine learning algorithm can diagnose the thermal imaging fault of brushless DC motor ventilation in enterprises, to improve the business's ability to resist the operation risk [7]. Aiming at the operation failure problem of electric percussion drills in the supply chain of manufacturing enterprises, the binary image diference common area (BCAoID) method is used for feature selection. Ten, the nearest neighbor classifer and back propagation neural network are used to analyze the data after feature extraction. It is found that this method has good accuracy, can efectively detect the failure problems faced by enterprises in the process of application, and can help enterprises to make early decisions on the method risk [8].

Research Progress on Prevention of Digital Financial Risk in Supply Chain.
With the universal of informative Internet science and technology, the management model of "Internet plus Finance" is requisite. It uses the advantages of advanced technologies such as mobile communication and blockchain to solve the fnancial management problems in the supply chain, strengthen the monetary afairs administrator in the operation process of enterprises, and decrease the economic afairs risks. To evaluate the threats brought by transactional fnance in the supply chain, a machine learning algorithm can form a matrix for forecast analysis [5]. Te average merge method can establish a blend dominate mold, and decision-making is made using algorithms such as the decision tree model, gradient promotion model, and random forest model. Te operation status of the enterprise is analyzed, and risk prediction indicators are proposed to continuously optimize the demonstration of the risk dominant models, to raise the natural capacity of the enterprise to resist monetary afairs risks [9]. For the network credit fnancial risks in the supply chain, a corresponding risk early warning mechanism should be established. Te P2P peer-to-peer lending model is a typical model of the Internet fnance industry [10]. Analyzing and researching its risks is helpful to fnd out the risk factors in the whole of Internet fnance. Te weighted KNN algorithm lies in volatile accuracy, and the coarseness set can control the Internet fnancial risk. Using the relative superiority and inferiority resembling notion of diverse accuracy and coarseness set, diferent training sets are divided into a central region and boundary region. According to the matching of the sample center and test sample, the area to which the selection belongs is obtained, and the category of the part to which the instance belongs is determined. At the same time, the risk level of each sample can be determined by a quantitative weighted KNN algorithm, which is benefcial for managers to make scientifc and reasonable decisions [11]. For the fnancial risks of capital operation in the supply chain, it can also be designed by using the web application program in the network embedded system. Using the embedded design, it can realize the online operation risk control and prompt user behavior, embed the risk detection system of the fnancial software module in the daily decision-making process, predict the risks of market risks, trade credit risks, Internet risks, liquidity risks, etc., establish an early warning mechanism, and construct a multimonitoring and discriminant analysis model for risk early warning [12]. Following the requirements of relevant accounting standards, continuously improve the Internet accounting control system, track and warn online payment fees and accounting efciency in real-time, control accounts receivable ratio, shorten collection aging, improve fund management efciency, improve risk prevention capability, and control fnancial risks of enterprises [13].
For credit card fnancial risk in the supply chain, some scholars use an artifcial neural network to detect credit card fraud risk [14] and use supervised machine learning technology to classify credit card transaction data into fraud cases and accurate cases. Using an artifcial neural network to detect credit card fraud as a binary classifcation task, the digit set is segmented into 315 legal transactions and 492 fraudulent transactions. Te accuracy of the trained model can reach 99.95%. Te artifcial neural network model can well predict credit card fraud transactions, help managers make decisions in advance, and prevent fnancial risks [15]. Tere are also some relevant scholars who use the diferential evolution superparameter optimization method to detect credit card fraud, use the diferential evolution algorithm to deal with the data imbalance problem, and use the optimized XG-Boost algorithm to classify fraudulent transactions. Te model has high accuracy after evaluation [16]. Some scholars also analyze the detection performance of fraudulent transactions based on a meta-heuristic algorithm and use meta-heuristic technology to optimize the superparameters, which can validly enhance the usefulness of fraud examination systems, simplify the detection process, and shorten the detection time [17]. Te genetic algorithms can also refne the super specifcations of fraudulent transactions, and genetic algorithms and network search algorithms are compared and analyzed to compare the accuracy of credit card fraudulent transactions experiments. Te practical consequence shows that the genetic algorithm has higher prediction precision and better performance than decision tree, logistic regression, random forest, and other algorithms [18]. In addition, the particle swarm optimization algorithm can also refne the superparameters of the deep neural network to check credit card swindle trade. In contrast with the network search algorithm, the particle swarm optimization algorithm has better performance in precision, accuracy, recall rate, and other aspects, which can simplify the decision-making time and heighten the usefulness and efectiveness of fraud transaction detection [19].
As for the fnancial risks of bank loans, fnancial institutions are vulnerable to fraudsters and are exposed to fraud risks. Artifcial intelligence technology helps to collect and sort out data, collect and evaluate credit information of lenders, etc. It can reduce the risk of default of lenders faced by banks [20]. Still, too excessive risk prevention will also cause some potential benefciaries to lose the chance of loans, which is not conducive to the popularizing of national fscal policies. Terefore, the risk prevention coefcient of bank loans should be moderate, and the efciency of loan management needs to be improved urgently [21]. Most banks adopt a labor-intensive management approach, but it is less efective and will cause many people to lose their jobs. Te traditional management approach has been unable to fully satisfy the requirement of monetary afairs organizations fully. It obliges make use of artifcial intelligence, machine learning, and other models to enhance the risk prevention capabilities of fnancial institutions and improve their governance level [22]. Artifcial neural network is used to examine the swindle risk in bank advance administration, which not only restricts the behavior of loan benefciaries but also monitors loan fraud. Selecting the loan credit data of more than 600 customers of a bank, extracting the features, and then predicting and making decisions, the precision of the mold can achieve 98%. Tis model can efectively help fnancial service institutions to detect credit risk and prevent the harm caused by fraud [23].
For supply chains in diferent industries, machine learning algorithms can also solve the fnancial risk problem of business [24]. For the energy supply chain, we can not limit ourselves to the traditional regression method [25] but use a meta-elastic network learner to constitute the data of various learners, to achieve the purpose of prediction and decision-making. By adding additional energy prediction indicators for the industry to the decision-making data, potential risk information faced by enterprises can be predicted more pertinently, and the accuracy of the fnal decision can be improved [26]. For the supply chain of the construction company, it is particularly imperative to distinguish the risks of the enterprise and take corresponding measures to control various hazards [27]. Cross-analysismachine learning model can be used to predict the chances of large-scale construction projects in the construction industry. By obtaining from the company the construction cost, completion time, quality control, impact scope, and other factors that will afect the operation risk of the enterprise of large-scale construction projects, and obtaining from the experts the Likert scale after scoring each risk factor, the received data are further processed, mainly for the identifcation of high-risk components and relevant low-risk elements, and performing K-means clustering analysis to get features with apparent characteristics. Firstly, descriptive statistics are carried out to analyze each data type; then, a few excessive-sampling skills (SMOTE) and Wilcoxon rank and sum test are synthesized, which helps to retain essential Computational Intelligence and Neuroscience eigenvectors such as construction cost, completion time, and quality control. Finally, the genetic algorithm-based Kmeans clustering analysis algorithm (GA-K-means) employs the biobjective gamma to break up the extreme risk elements and the related low-risk factors, thus helping companies to determine diferent risk levels and take corresponding measures promptly [28].

Financial Risk Prevention Model
As can be seen from the relevant research foundation, for fnancial risk prevention, most scholars use the neural network algorithm to study the relevant content, and few scholars apply the chaotic grasshopper algorithm to the supply chain fnancial risk prevention model and introduce the slime mold optimization algorithm to optimize the superparameters in the classifcation process. Te workfow of the algorithm of the supply chain digital fnancial risk prevention model suggested is shown in Figure 1. First, collect relevant fnancial data of supply chain enterprises and input these data into the training set as training data; secondly, the data are preprocessed, and the CGOA algorithm is selected to select the features of these data. Putting the chosen functions into the training data set, and establishing a part of test data to confrm the precision of the pattern at a later stage; then, the classifcation process is carried out, SMA is used for parameter adjustment, and the SVM algorithm is used for data classifcation. Finally, these data are added to the trained model, and the precision of the mold operation is evaluated.

CGOA Based on Feature Selection.
Grasshopper optimization algorithm (GOA) was initially proposed by Saremi et al. [29] in 2017, and Scholars Fouad [30], Ukasik [31], and Meraihi et al. [32] improved the algorithm, respectively. As a meta-heuristic bionics optimization algorithm, its search usefulness and convergence capability are better. Its unique adaptive mechanism can better bring into equilibrium the general situation and local search process, thus improving the search accuracy. Te locust optimization algorithm simulates the removing and fnding foodstuf behavior of the locust population in the natural world and divides the whole process into an exploration process and a development process [33]. Te algorithm is principally a compound of three factors, and its scheme process is shown in the following formula: Here, X i indicates the position of the i-th locust, S i expresses each other attraction between individual operations, G i indicates the gravitational force of the i-th locus, and A i means the wind intensity received by the i-th locust. Te location of locusts is most afected by social action, and the formula for calculating S i is (2) Here, N represents the total quantity of locusts in the community, d ij indicates the range between locust i and locust j, and S is the function of social strength. Te estimation process is as follows: Here, f and l are the social attractiveness strength and social attractiveness scale, respectively, and r represents the actual social value of its existence.
Secondly, the formulas of G i and A i are Here, g is the gravitational force target, e g is the normalized vector facing the earth, u is the wind literal, and e w is the normalized vector direct to the direction of the wind. Terefore, the position X i of the locust can be expressed as However, in this mathematical model, the locusts will quickly fnd a comfortable position, and the population will not converge to a specifc value, requiring optimization. At this time, the gravity factor G i is not considered, and the wind direction always points to T d , so the improved mathematical model after optimization is Here, ub d and lb d are the superior and inferior bounds of S(r) in the d-dimensional room, individually, and T d is the optimal solution of the locust location in the d-dimensional room. c is the decreasing coefcient, and the calculation formula is Here, c max and c min are the biggest and smallest number of parameters c; individually, n is the times of iterations present, and n max is the biggest times of iterations.
Te population initialization of the traditional GOA algorithm is entirely random, which cannot guarantee the ergodicity of the population in the sample space, and the optimization accuracy is poor. Terefore, this paper 6 Computational Intelligence and Neuroscience introduces the initialization method of chaotic mapping on this basis, using Tent mapping to obtain [0, 1] constant value in the space. Its expression is In the formulation, i is the population size, i � 1, 2, · · · N, j � 1, 2, · · · d , and j is the chaotic serial number, which represents the spatial dimension of the individual. Take the random number of [0, 1], determine the chaotic parameter of [0, 2], take the initial value of formula (8) to obtain the y chaotic sequence y i,j , and then use formula (9) to perform chaotic mapping: In the formula, [l j, min , u j, max ] represents the search range of x i,j .
Te adjustment formula for parameter c is In the formula, l and L are the times of iterations present and the biggest times of iterations, individually and c shows nonlinear iterative decrement. In the initial phase of iteration, c is enormous, and the decrement velocity is sluggish, which can seek the population data; in the posterior period, c is small, and the decline rate is faster, so it can quickly converge with good convergence efect.
In addition, CGOA takes into account the leading role of elite individuals T d ′ ∧ ′ on the position of the population in the iterative process and introduces a probabilistic perturbation strategy to the current elite individuals, and the perturbation probability P dis is

Computational Intelligence and Neuroscience
Te perturbation method is stochastic and produces a random digit r between [0, 1]. If r > P dis , individual perturbation is required. If r ≤ P dis , the original individual remains unchanged.
At the same time, the Cauchy variation mechanism is introduced. Te Cauchy distribution is uninterrupted, and it represents probability [34]. Te probability density function of the one-dimensional is Introducing the Cauchy operator into the position formula of the current optimal solution, we obtain the following equation: In the formula, Cauchy is the Cauchy operator obeying the Cauchy distribution.
Combined with the disturbance probability, the current optimal position is In short, the overall execution process of the CGOA algorithm is revealed in Figure 2.
CGOA algorithm can be used to select the features of the existing data and set diferent warning values for the selected elements. If it does not exceed the warning value, it indicates that the relevant indicators can operate generally within a reasonable range of changes; if this warning value is exceeded, it suggests that there is fnancial risk. Te supply chain fnancial risk prevention model will give the warning to inform managers that decisions should be suspended to solve the fnancial risk problems faced promptly and prevent enterprises from expanding losses.

Support Vector Machine Algorithm
Structure. Support vector machine (SVM) is a broad linear catalog that performs a duality catalog on data by supervised learning [35]. It applies the hinge loss function to compute the operation experience risks. It includes a holomorphic item to the solution setup, to achieve the purpose of optimizing the structural risk. SVM can use the nonlinear mapping principle: f: R n ⟶ H, to shine upon the nonlinear problem in a high-dimensional room, thereby transforming it into a linear problem [36]. Te conforming sample data set is D � x i , y i , i � 1, 2, . . . n, x i ∈ R n , y i ∈ R, where x i is the aferent feature eigenvector and y i is the output load value; the SVM algorithm transforms x i into a high-dimensional room by shining upon f(x) to the expression: In the formula, ω is the slope and b is the intercept. x � (x 1 , x 2 · · · x n ) is the feature vector, and y � (y 1 , y 2 · · · y n ) is the output load value.

Computational Intelligence and Neuroscience
To fnd the optimal ω and b, the expressions need to satisfy the following conditions: In the formula, C is the punishment term, ξ i and ξ * i are the relaxation term, having some error, and ε is the most signifcant deviation, which is the parameter of the linear insensitive loss function.
To enhance the overall optimization efect of the model, the Lagrange multiplier method can translate it into a twofold matter: In the formula, a i and a * i are Lagrange multipliers, K(x i , x j ) is the kernel gamma, and its expression is In the formula, c is the kernel parameter. So the fnal expression is In the support vector machine (SVM) algorithm, the punishment term C and the kernel parameter c are momentous throughout the whole course. C refects the generalization ability and error size of the pattern. Te more signifcant the C is, the smaller the efect of the pattern. However, the smaller the C is, the smaller the generalization ability of the pattern. c will afect the ft of the model. To enhance the model's accuracy, it is requisite to fnd a better combination of (C, c), so it needs to be further optimized. Te calculation fow of the Support Vector Machine is shown in Figure 3.
Te SVM algorithm can categorize the specimen data after feature selection. Based on diferent early warning indicators, diferent risk levels can be classifed. Te results after classifcation are helpful for supply chain managers to timely perceive the current operating conditions and take corresponding measures to prevent supply chain fnancial risks.

Support Vector Machine Algorithm Based on SMA
Optimization. Among the swarm intelligence algorithms, the slime mould algorithm (SMA) was put forward by Li et al. [37] and other scholars in 2020. It mainly simulates the spread activity and looks for the food behavior of slime molds [38]. Te adaptive weight afects the propagation wave of slime molds on the basis of organism oscillators, thereby generating positive and inverse feedbacks, which is an optimal connection path with better exploration ability and development tendency [39]. Te realization process is mainly split into the following measures.
Firstly, initializing parameters and the position of a population, wherein the population satisfes the characteristics of randomly distributed search individuals in a search space, the formula is Medium formula, X → means the location of the slime mold, t means the present iteration times, X A � �→ and X B � �→ represent the individuals randomly selected from the slime mold, respectively, tan h represents the hyperbolic tangent function, and W �→ represents the probability of slime mold, vb → and vc → both represent parameters that meet a specifc range, and the expression is In the formulation, T is the time of iteration at present, Max t represents the most considerable iteration times, and artanh represents the inverse hyperbolic function.
In the second phase, the adaptive value of apiece search individual is counted, and then the search individuals are sorted, and the weight W �→ is calculated as Computational Intelligence and Neuroscience In the formula, i � 1, 2, · · · N, j � 1, 2, · · · D, N is the population number, and D is the dimension. r represents a stochastic digit in the domain of [0, 1], and bF and wF represent the frst-rate ftness and the worst adaptation embodied in the current iteration process, individually. sort(F) means the result of sorting the ftness values.
Te third step is to fnd the overall situation's best place and ftness value. To get the individual's current optimal position X → , continuously adjust the parameters vb → and vc → , and the weight W �→ , draw the individual's activity trajectory in the three-dimensional space, and seek out the best consequence, the expression formula is In the formula, X i means the i-th search individually, rand refers to a random digit between the values [0, 1], and the UB and LB distributions embody the superjacent and nether demarcations in the seek area.
Te fourth step is to update the individual location. After adjusting the parameters vb → and vc → and the weight W �→ , the location expression of the scouted individual operation is updated as In the formula, both rand and r represent random values between [0, 1].
Te overall operation course of the slime mould algorithm is revealed in Figure 4.
To enhance the classifcation efciency of the support vector machine algorithm and explore the potential development ability of SVM, this paper proposes an optimized support vector machine algorithm based on the slime mould algorithm (SVM-SMA). Te foraging algorithm adjusts and continuously optimizes the relevant parameters to enhance the rationality of the fnal decision.
In the SVM-SMA model, through the optimization of the slime mould algorithm, the situation of the searched private ownership is adjusted, and the support vector machine algorithm is used to calculate the ftness of the searched individual, and the following parameters are used for classifcation, to achieve the purpose of optimizing SVM: To sum up, the execution fow of SVM-SMA proposed in this paper is revealed in Figure 5.

Numerical Examples
In this paper, all A-share listed companies of China Shenzhen Stock Exchange and Shanghai Stock Exchange in recent three years are selected as research samples. Te data are from stock exchanges. Firstly, the data are preprocessed, and the screening process includes the following procedures: (1) eliminating companies with a shortcoming or exceptional data; (2) standardization processing of sample data; (3) excluding the fnancial industry. Ten, the panel data are analyzed with STATA13.0, and all continuous variables are subjected to 1% up and down Winsorize to overcome the infuence of extreme values on the experimental results.
Python is used to model the preprocessed data. Firstly, the chaotic locust optimization algorithm is used to select the features. After selecting the elements, the range of the data is simply analyzed. Descriptive index data are expressed  Table 1.
From the data index, we can discover that the amount of companies increases gradually over the years, showing an increasing trend. Each characteristic value will also have a slight change in diferent years, but the diference is not signifcant and is relatively stable as a whole. Each characteristic of the same year will be other, and each character is representative to a certain extent.
In view of the existing research, this paper constructs a confusion matrix to examine the efect of the decision of strategic importance of the pattern [40], examine the relationship between the actual results and the decisionmaking results, and explore the decision-making accuracy of the pattern. Te confusion matrix is displayed in Table 2.
If the actual result is normal and the decision result is normal, then TP; if the true result is normal and the decision result is abnormal, FN; if the true result is abnormal, and the decision result is normal, FP; if the true result is abnormal and the decision result is abnormal, it is TN. TN and TP represent the number of correct decisions made by abnormal samples and normal samples, respectively. Te higher the value is, the better the decision. FN and FP represent the number of erroneous decisions with abnormal samples and those with normal samples, respectively. A smaller value is better, indicating a smaller error.
Due to the extensive data of the sample companies selected in this paper, the decision-making results cannot be seen more intuitively only based on the confusion matrix. Terefore, relevant evaluation indicators are constructed on this basis, and the model results are evaluated through the calculation of indicators. Te specifc indicators are shown in Table 3. In this paper, the CGOA is applied to feature selection, the SVM is used to classify the data after feature selection, the SMA is used to adjust the parameters to seek out the best solution, and the fnancial risk standard of the supply chain is evaluated. Te average value of each year is calculated based on the results, and the results are compared with the decision tree model and the BP-NN model. Te decision consequences are displayed in Table 4.
From the results, for overall accuracy, the model CGOA-SVM-SMA proposed in this paper has a better decision-making efect. In the general sample, the proportion of correct decision-making is 85.38%, which is higher than the comparison of the two models. For F-Score and TNR, which are two harmonic averages, the accuracy rate has decreased, because the sample companies have diferent accuracy rates in normal samples and abnormal samples. However, in general, CGOA-SVM-SMA is better than else patterns in three aspects of PRE, F-Score, and TNR and has higher accuracy. Te application of the pattern in fnancial risk prevention of enterprises has specifc practical value.

Conclusion
Today's Internet technology has become more and more popular, and the Internet fnancial risks of supply chains can be seen everywhere, bringing enormous challenges to the development of enterprises. In order to help supply chain enterprises make better decisions and deal with the fnancial risks faced by enterprises, this paper proposes a supply chain digital fnancial risk prevention model, which mainly includes the pretreatment stage, the feature selection stage based on CGOA, the data classifcation based on SVM, and the parameter optimization process based on SMA. Preprocessing the data in the light of collecting the data to improve information quality; in the feature selection stage, the chaotic grasshopper optimization algorithm is further optimized based on the grasshopper optimization algorithm, and the preprocessed data indexes are selected for feature selection, which paves the way for the following research; the robust support vector machine algorithm is used to classify the data indexes. Based on that, the slime mold foraging optimization algorithm optimizes the parameters to fnd the best solution for the decision system. Using this model to analyze the relevant data in the recent three years, it is found that the simulation results obtained are efective in fnancial performance, accuracy, sensitivity, and other aspects and have a good application prospect, which can efectively and accurately help supply chain enterprises to prevent fnancial risks. Trough the model validation of the sample data of all    A-share public corporations in China Shenzhen Stock Exchange and Shanghai Stock Exchange, it is found that the model can well prevent fnancial risks and enhance the fnancial performance result of the supply chain. Te accuracy rates of the data model can reach 85.38%, respectively, after the test, indicating that the model has good prediction and decision-making efects and can enhance the reliability and reliability of the decision-making results. Te main body of the supply chain can discover the fnancial risks faced by the enterprises in time with the aid of the model, analyze the risk levels faced by the enterprises, and take corresponding measures to prevent the fnancial risks in the supply chain in time. Te limitation of this research lies in that the infuence of outliers is not considered, the companies with the abnormal operation and missing data are removed in the data preprocessing process, and only representative extensive sample data are selected. In future research, the method of outlier detection can be added to the model CGOA-SVM-SMA or the model can be improved to enhance further the precision of the supply chain fnancial risk prevention pattern.

Data Availability
Te data used to support the fndings of this study are available from the corresponding author upon request.

Conflicts of Interest
Te authors declare that they have no conficts of interest.