Credit Risk Evaluation of Asset Securitization of PPP Project of Sports Public Service Venues Based on Random Forest Algorithm

Due to the characteristics of sports public service venues, there are still some financing difficulties under the PPP (public-private partnership) operation mode. Although asset securitization can solve its corresponding problems and enhance the standardization of PPP projects, its many participants, complex transaction structure, and other influencing factors still have a great impact on the financing effect. Therefore, this paper integrates the random forest algorithm into the asset securitization credit risk evaluation system of PPP project of sports public service venues, and screens the audition indicators of credit risk evaluation through the constructed model. The experimental results show that optimizing the parameter setting of random forest model can effectively reduce the misjudgment rate of default samples, while maintaining the stability of other performance indicators. Even if other misjudgment rates will increase to a certain extent, it is within a reasonable range, which improves the overall performance of the model. The ROC curve shows that the risk credit indicators selected by the model have strong evaluation performance and effectiveness and can provide some reference information for the credit risk prevention of asset securitization of PPP project of sports public service venues.


Introduction
e improvement of national fitness awareness is one of the important driving forces for the development of sports industry. More different types of sports have gradually entered public life. People's demand for sports venues has also changed from simplification to diversification and specialization. e construction of sports public service venues has entered a stage of rapid development [1]. For any country or city, hosting large-scale sports events, especially international sports events, can promote economic development, promote the adjustment of economic structure and the solution of social employment and other problems, and improve the popularity of countries and cities. Similarly, countries and cities that can host large-scale sports events also show certain advantages in terms of infrastructure, environmental conditions, comprehensive strength, and event experience [2]. However, with the development of sports events, the infrastructure and stadiums established in the past cannot meet the needs of more and more professional sports events. e stadiums and gymnasiums operated by the government after the game cannot produce good social benefits and even become idle buildings because of the single function, low actual utilization rate and the operation mode do not meet the market demand. is situation undoubtedly increases the financial burden of the local government and fails to achieve the purpose of making full use of social resources [3]. erefore, combined with past cases and experience, the construction and management of sports public service venues has produced a form of cooperation between the government and social capital, namely PPP mode. With the participation of the government, it actively absorbs social capital, separates the ownership and management right of sports public service venues, and produces various types of cooperation modes [4]. erefore, the research in this paper study PPP project, it as a new financing mode of sports public service venues, PPP mode has obvious effect in reducing the financial pressure of the government in the early stage of development and promoted the development process of public utilities. On the other hand, during the development of PPP projects, problems such as difficult financing and difficult withdrawal of social capital have gradually emerged. ese problems existing in the PPP project of sports public service venues can be solved to a great extent by means of asset securitization, which is also the new direction and new road of its development.
Basic assets are an important foundation and guarantee for the realization of asset securitization of PPP projects. Asset securitization can improve the standardization of PPP projects and reduce the difficulty of financing [5]. However, the PPP project of sports public service venues itself has a long life cycle, many participants, and complex transaction relations.
erefore, there are many risk factors that will affect the asset securitization of the project [6]. e credit of assets is the foundation for PPP projects to realize asset securitization and promote project development. Based on the random forest algorithm, this paper constructs the credit risk evaluation model and the corresponding credit risk evaluation index system of asset securitization of PPP project of sports public service venues, verifies the evaluation model through empirical analysis, and analyzes the results.

Research Status of Asset Securitization of PPP Projects at Home and Abroad
PPP mode was initially emerging abroad. Many scholars have studied the risk and possibility of projects operated through this mode. Some scholars have analyzed the risktaking mechanism of PPP model from the perspective of macrosystem, microsystem, and private sector, and pointed out that there are differences in risk types of different PPP projects [7]. As for the necessary conditions for asset securitization to achieve the purpose of financing, some scholars believe that this method can reduce the financing cost of sponsors and stimulate the enthusiasm of investors and the degree of market participation [8]. Some scholars also pointed out that asset securitization is the process of transforming fixed assets into securities, and the process of transforming illiquid assets into valuable investment, which can promote market liquidity. Similarly, it has a certain demand for the market environment [9]. erefore, for the asset securitization of PPP projects, some scholars pointed out that asset securitization is an important means to improve the liquidity and profitability of PPP projects [10]. Other scholars have concluded through corresponding research that PPP projects can effectively reduce financing costs through asset securitization, and its financing effect will be affected by environmental factors such as government and public support [11]. Relatively speaking, the development of domestic PPP mode started late and there are relatively few examples that can be studied. e research in many aspects, such as project feasibility, impact and prevention of risk factors, is still in the exploratory stage [12]. Some scholars concluded through analysis that the risks of PPP project itself, the risks of all participants and the risk of benefit distribution need to be analyzed and prevented. At the same time, the project contract risks, especially the terms of franchise rights, need to be paid attention to [13]. Some scholars studied the credit enhancement system of asset securitization and pointed out that the effect of the current credit enhancement system in reducing credit risk is not ideal, and the system needs further improvement [14]. Other scholars put forward suggestions beneficial to the progress of asset securitization of domestic PPP projects based on the WBS financing model experience of Britain and the United States [15]. According to the domestic urban construction process, some scholars believe that the development of asset securitization of PPP project meets the needs of urban construction and development. It can meet the needs of a large number of infrastructure and public construction, relieve the pressure of government funds, reduce project financing risks, and broaden financing channels [16].

Feasibility of Asset Securitization of PPP Project of Sports
Public Service Venues. Firstly, the realization of asset securitization of PPP project of sports public service venues needs the support of relevant policies. In the released policy document, it is pointed out that while reducing the high debt ratio of state-owned enterprises, it is necessary to provide good policies and create a good institutional environment, encourage the absorption and supplement of social capital according to law, broaden the channels of equity financing, inject new capital vitality into the stock assets of state-owned enterprises, maximize the benefits of existing resources, and provide a strong driving force for the virtuous circle development of sports industry [17]. By virtue of the communication and cooperation mechanism, diversified investment and financing subjects can increase communication channels and obtain more relevant information, reduce the operation risk of sports public service venues, improve their operation efficiency, and promote the process of asset securitization of the project [18]. Secondly, social capital provides the original capital for the asset securitization transformation of PPP projects of sports public service venues. Social capital includes idle funds owned by residents and large-scale savings funds, and their huge investment benefits have not been brought into full play [19]. Sports public service venues need capital accumulation, and the existence of social capital provides them with the possibility and conditions to complete capital accumulation [20].
Finally, the asset securitization transformation of PPP project of sports public service venues meets the market demand. Project asset securitization effectively reduces the dependence of stadiums and gymnasiums on government and state-owned funds. At the same time, sports public service stadiums and gymnasiums have good reputation and are guaranteed by future income and government tax, which increases their securitization credit and reduces investment risk [21]. e process of project asset securitization is also the embodiment of realizing the economic and social benefits of venues, which can play its potential functional value [22].
As shown in Figure 1, it is the asset securitization process and corresponding subjects of PPP project of sports public service venues.

Audition and Data Processing of Credit Risk Assessment Index System for Asset Securitization of PPP Project.
Compared with the general project asset securitization, the PPP project asset securitization of sports public service venues has a long cycle, complex links, and diversified participants. A credit crisis in any link will damage the securitization quality of the whole project. erefore, the credit risk of project asset securitization is the most concerned link of investment participants. e audition of the evaluation index system is carried out according to the characteristics of sports public service venues, the credit indicators of Galway at home and abroad, and the index system sorted out in relevant literature research. Its credit risk mainly comes from the credit risk of each participant in the process of securitization, government credit risk, basic asset credit risk of PPP project of sports public service venues, and transaction structure credit risk, which is used as the criterion level to build an audition index set [23].
In order to ensure the standardization and unity of a large amount of relevant index data, data preprocessing measures need to be taken to prepare for the screening of indicators in random forest model. e data standardization processing is shown in equations (1)- (3).
Standardized equation of positive credit evaluation index: Standardized equation of negative credit evaluation index: Standardized equation of interval credit evaluation index:

Principle of Credit Risk Assessment System for Asset Securitization of PPP Project Based on Stochastic Forest
Algorithm. e unit classifier of random forest algorithm is cart classification tree and contains multiple classification trees. Each classification tree corresponds to a training subset formed by independent sampling and maintains the same index selection probability, which is repeated constantly [24]. e results of these classification trees are different and independent of each other, and then the final classification results are selected by simple voting on the results of the classification tree [25]. As shown in Figure 2, the establishment flowchart of index system based on random forest algorithm is shown. e cart classification tree in the random forest model is like a tree, which contains three nodes: root, middle, and leaf. e root node is the place where all sample data are input. e input samples have the same number of index attributes, and these index attributes form a candidate index set to be divided. After traversing each index attribute, there is and only one selection opportunity. In the middle part of the node, the optimal classification index and corresponding value selected in the sample are divided according to the Gini coefficient principle, and the cycle operation is carried out until all the samples contained in the leaf node belong to the same class. At this time, the candidate index set is empty, and if it is satisfied at the same time, the segmentation stops. Let T represent the node before segmentation, t 1 , t 2 is the node after segmentation, the classification type represents Y, the number of sample types contained in T is j, and the probability of a category in the node after segmentation is represented by P(Y n | t m ). When the split point is selected as a specific value in the index, the Gini index of the split node t m is calculated as shown in the following: e above indicates that when the segmentation node contains only the same type of samples, the Gini index reaches the minimum value. When the segmentation node contains different types of samples with the same proportion, the Gini index reaches the maximum.
e Gini impure of any index attribute divided by the node can be calculated by where the number of split nodes is m, the number of samples in the split node is i m , and i represents the total number of samples in the node. e independent sampling of cart classification tree is a random sampling operation with return, so the probability Computational Intelligence and Neuroscience of each sampling is equal to each sample, all 1/i, and the i probability of not being selected in the second sampling is shown in the following: If the number of original samples in the above equation is sufficient, the probability value will gradually converge. e convergence value is 1/e, that is, only nearly 63.2% of the samples in the sample set may exist in the training subset, and the nonexistent part belongs to "out-of-bag data," that is, OOB.
After the construction of random forest model is completed, its prediction estimation is completed through OOB, and its accuracy estimation is shown in the following: e denominator in the above equation is the total number of samples that have not been drawn, and the numerator is the number of correctly classified samples that have not been drawn. e prediction result of each classification tree represents a type vote, and the category with the highest number of votes is the prediction classification label. For example, (8) represents the calculation of prediction result: where ntree represents the number of decision trees and δ(•) represents the indicative function. e generalization error obtained from the OOB input model can effectively measure the classification performance of the model. e interval function and the mathematical expression of generalization error are shown in the following: Pe * � P x,y (mg(x, y) < 0).
In the equation, the average value of the sample correctly predicted with label y is expressed as ave k δ(h k (x) � y), and the maximum average value of the sample not predicted with label y is expressed as maxave k δ(h k (x) � n) n≠y ; Pe * is the generalization error.
In the form of interval function, its edge function can be defined as shown in the following: In the equation, the probability represented by P(h m (x) � y) is the probability of correct judgment, and max n≠y P(h m (x) � n) is the maximum probability of incorrect judgment.
As shown in Figure 3, the confusion matrix of the evaluation index of random forest algorithm is divided into two categories: nondefault samples, i.e., 0. Otherwise, it is    FN) .
e specificity calculation equation is shown in the equation: e accuracy is calculated by: e accuracy calculation is shown in the following: ACC � (TP + TN) (TP + FN + TN + FP) . (15) e error rate calculation equation is: By introducing the influence of random noise into the random forest model, the importance level of credit evaluation indicators can be determined by the change of prediction accuracy. e calculation equation is shown in the following: where ACC km represents the OOB estimation accuracy of k classification tree after all out-of-bag sample values of index attribute m in random order. e weight value of credit risk evaluation index is calculated according to e weight value is expressed as w m and the number of credit risk assessment indicators of asset securitization is expressed as M.

Computational Intelligence and Neuroscience
On this basis, the index importance level and its relationship of random forest algorithm are reflected by where E(C) m represents the credit evaluation score, y (mn) is the standard value of the n credit evaluation index of m, and w n is the weight value. e credit rating range (as shown in (1) and (20)) needs to be converted to the credit rating range of [1,20]:

Empirical Analysis on Credit Risk Assessment of Asset Securitization of PPP Project of Sports Public Service Venues Based on Random Forest Algorithm
e contract term of a large-scale sports public service venue PPP project is 25 years. According to the relevant contracts, the project company is responsible for the construction of sports public service venues, has the right to operate sports public service venues, and is responsible for their operation, daily maintenance, venue transformation, and renewal. During the contract period, the venue shall provide the government with event services that meet the event standards, and the project company shall ensure the integrity and good operation of the stadium. According to the characteristics of the project, the audition indicators of asset securitization credit risk of the project were shown in Table 1.
After data preprocessing and standardization, 70% of the credit evaluation samples are randomly selected as the training set and the rest are the test set. Random forest algorithm will produce different results due to the difference of many parameter settings. In order to reduce the subjectivity of parameter settings, it will be analyzed through two aspects: model selection characteristics and decision tree. As shown in Figure 4, the relationship between the number of different decision trees and OOB error when the range of feature numbers is determined. It can be concluded that the model OOB error does not change monotonically with the increase of the number of selected features, but shows a fluctuating state. Similarly, the OOB error does not decrease with the increase of the number of decision trees. Relatively speaking, when the number of selection features is four and the number of decision trees is 100, the OOB error appears local minimum.
In this case, let the maximum feature number be four, and analyze the OOB error of the number of decision trees from 80 to 120. As shown in Figure 5, when the number of decision trees is 105, the OOB error value reaches the minimum.
As shown in Figure 6, it is a scatter diagram between the mtry parameter of the model and the OOB misjudgment rate of the model. e larger the mtry parameter, the greater the OOB misjudgment rate. When it reaches 38, the misjudgment rate tends to converge and fluctuates in a small range. erefore, the optimal value of the parameter mtry is 1.
In order to confirm the rationality of parameter setting and model optimization, the judgment results of the model with default parameters and optimized parameters are compared, and the results are shown in Figure 7. e comparison results show that under the default parameters, the misjudgment rate of default samples is higher, and the misjudgment rate can be effectively reduced by optimizing the parameter settings. While other evaluation indicators remain in a high state after optimization, and the increase of some indicators is also within a reasonable range, so the optimization of model parameters is reasonable.  X21  X27  X22  X4  X32  X3  X2  X33  X8  X6  X5  X16 X19 X15 X26 X35 X12 Figure 9: Final credit risk evaluation index and its importance distribution.

Computational Intelligence and Neuroscience
Further verify the effectiveness of the risk assessment module analysis results, and randomly select samples different from the original sample set for judgment, as shown in Figure 8. e data showed that the model evaluation index results obtained after the optimization of the randomly selected sample set are high, and the misjudgment rate remains low, which meets the requirements of index screening.
Keep one of the repeated indicators in the audition indicators, and then randomly select three different sample sets for the same parameter setting optimization to screen the remaining credit risk evaluation indicators. e indicators that appear repeatedly in the model results are selected as the final credit risk evaluation indicators. Figure 9 shows the final credit risk evaluation indicators and their importance distribution. e results show that the effectiveness of the measures of bankruptcy isolation in the process of asset securitization of the PPP project of sports public service venues is weak.
As shown in Figure 10, the ROC curve of the random forest model in this paper can reflect the effectiveness of the model. ere is a large area below the curve in the figure, which means that the credit risk index system in this paper has strong evaluation performance and high effectiveness for the credit risk of asset securitization of PPP project of sports public service venues.

Conclusion
In the previous years, sports public service venues were mainly funded and operated and managed by the government. However, because the management mode could not keep up with the market demand, the single venue function could not meet the diversified needs of the public and many venues could not operate normally, while increasing the financial burden. PPP mode could use social funds to alleviate the pressure of government funds, improve the operation and management ability of venues, and follow the development of market demand. However, this model was still in its infancy, the system standardization was relatively low, and there were still many problems. Asset securitization can improve the standardization of PPP projects, reduce financing costs, and provide more guarantee for the development of sports public service venues. is paper introduces the stochastic forest model into the credit risk system of asset securitization of PPP project, and constructs the corresponding evaluation index. e experimental results show that the credit risk evaluation model based on random forest algorithm can maintain high model performance indicators, reduce the misjudgment rate of default samples, and improve the overall performance of the model. e credit risk indicators selected by the model have strong evaluation performance and high effectiveness for the credit risk of asset securitization of PPP project of sports public service venues. e random forest model constructed in this paper still has shortcomings and its performance in the extended data set needs to be verified. At the same time, the process of asset securitization of PPP project of sports public service venues involves many links. e influencing factors studied in this paper are limited and need to be further expanded.

Data Availability
All data used to support the findings of this study are included within this article.

Conflicts of Interest
e author declares that there are no conflicts of interest.