Decision tree analysis-based staged development evaluation method for entrepreneurship incubation project

Abstract

The invention discloses a decision tree analysis-based staged development evaluation method for an entrepreneurship incubation project. The method comprises the following steps of S1, collecting a multi-dimensional data set related to the entrepreneurship incubation project; s2, obtaining a preprocessed multi-dimensional data set; s3, forming a preferable evaluation feature set; s4, forming an initial evaluation model; s5, obtaining an optimization evaluation model; s6, generating sensitivity analysis results of the key risk points; and S7, inputting the real-time multi-dimensional data of the to-be-evaluated entrepreneurial incubation project into the risk sensitivity analysis module, calculating a stage development evaluation result of the entrepreneurial incubation project based on the optimization evaluation model, and identifying key risk points of development of the entrepreneurial incubation project according to the output of the risk sensitivity analysis module. And generating an evaluation report including the success probability, the potential risk and the development bottleneck. According to the method, the performance difference of the project under the risk situation can be effectively identified, so that more reliable decision support is provided for an incubation mechanism.

Description

Technical Field

[0001] The present invention relates to the field of business incubation technology, and in particular to a phased development evaluation method for business incubation projects based on decision tree analysis. Background Art

[0002] With the rapid development of artificial intelligence and data mining technology, entrepreneurial incubation project management is gradually moving towards digitalization and intelligence. As an important part of supporting the development of start-ups, entrepreneurial incubation covers a complex process of resource allocation, project tracking, and phased evaluation. Accurately grasping the project development stage, identifying potential risks, and adjusting resource allocation strategies during the incubation process are the keys to improving the success rate of incubation. However, existing technologies still have many deficiencies in the phased development evaluation of incubation projects, and it is difficult to meet the needs of the current complex and changing entrepreneurial environment.

[0003] At present, the phased evaluation of entrepreneurial incubation projects usually relies on manual analysis by experts or simple single-indicator scoring methods. The existing methods usually score the single dimensions of the project's finance, market, team and product based on the experts' experience, or use a multi-indicator scoring model with fixed weights for comprehensive evaluation. Although it can provide a certain degree of reference, it has significant defects in the following aspects:

[0004] On the one hand, manual evaluation methods are highly dependent on expert experience and subjective judgment, making it difficult to form a unified evaluation standard, resulting in significant human errors in the evaluation results. When entrepreneurial projects involve multi-dimensional and complex features, manual analysis is difficult to fully capture the interactions between features, thus affecting the accuracy and comprehensiveness of the evaluation.

[0005] On the other hand, single-indicator scoring models usually use preset weights to conduct quantitative evaluation of projects. This fixed-weight method lacks the ability to dynamically adapt to the phased characteristics of entrepreneurial projects and cannot adjust the evaluation strategy in real time to reflect the changing needs of the project at different stages.

[0006] In recent years, machine learning technology has been gradually applied to the field of entrepreneurial incubation. Some methods attempt to use simple classification models or regression models to evaluate projects. However, existing methods mostly rely on shallow feature learning and cannot fully explore the complex relationships between multidimensional data. At the same time, such models show obvious lack of robustness when dealing with changes in the external environment, team fluctuations, and dynamic changes in market disturbances, resulting in evaluation results that are difficult to reflect the true project status. In addition, existing evaluation technologies are relatively weak in the interpretability and visualization of model results, making it difficult for entrepreneurs and incubators to formulate accurate resource allocation and optimization strategies based on the evaluation results.

[0007] To sum up, the existing technologies have obvious deficiencies in the accuracy of evaluation, dynamic adaptability, anti-interference ability and result interpretability. The above defects have seriously restricted the scientific management and efficient development of entrepreneurial incubation projects. To address the above problems, there is an urgent need for a new entrepreneurial incubation project phased development evaluation method that can comprehensively mine the correlation of multidimensional data, dynamically adjust the phased weights, enhance the anti-interference ability and have high interpretability, so as to solve the above defects of the existing technologies. Summary of the invention

[0008] One purpose of the present invention is to propose a phased development evaluation method for entrepreneurial incubation projects based on decision tree analysis, which can effectively identify the performance differences of projects under risk scenarios, thereby providing more reliable decision support for incubation institutions.

[0009] A method for evaluating the phased development of an entrepreneurial incubation project based on decision tree analysis according to an embodiment of the present invention comprises the following steps:

[0010] S1. Collect multi-dimensional data sets related to entrepreneurial incubation projects, including financial data, market data, team data and product data;

[0011] S2. Clean the collected multidimensional data set, remove noise data, fill in missing data, perform normalization, unify the data scales of different dimensional features, and obtain the preprocessed multidimensional data set;

[0012] S3. extracting a variety of evaluation features from the preprocessed multidimensional data set, optimizing the extracted evaluation features, and forming a set of preferred evaluation features;

[0013] S4. construct a staged evaluation model using the evaluation feature set, and train the staged evaluation model to form an initial evaluation model;

[0014] S5. Use the validation data set to validate the trained initial evaluation model, evaluate the accuracy and robustness of the initial evaluation model, adjust the model parameters according to the validation results, optimize the feature interaction ability and the predictive ability of the staged development of the initial evaluation model, and obtain the optimized evaluation model;

[0015] S6. Construct a risk sensitivity analysis module based on the optimization evaluation model. By optimizing the back propagation calculation of the evaluation model, identify the key evaluation features that have the greatest impact on the evaluation results during the phased development of the entrepreneurial incubation project, conduct quantitative analysis on the key evaluation features, and generate sensitivity analysis results for key risk points;

[0016] S7. Input the real-time multi-dimensional data of the entrepreneurial incubation project to be evaluated into the risk sensitivity analysis module, calculate the phased development evaluation results of the entrepreneurial incubation project based on the optimization evaluation model, identify the key risk points in the development of the entrepreneurial incubation project according to the output of the risk sensitivity analysis module, and generate an evaluation report including the probability of success, potential risks and development bottlenecks.

[0017] Optionally, the S1 includes the following steps:

[0018] S11. Collect financial data of entrepreneurial incubation projects and construct a financial dataset D describing the financial characteristics of entrepreneurial incubation projects f , including financial indicators of total revenue, total expenditure, cash flow and capital utilization;

[0019] S12. Collect market data of entrepreneurial incubation projects and construct a market data set D that describes the market characteristics of entrepreneurial incubation projects m , including market-related indicators such as market coverage, customer conversion rate, market share and market growth rate;

[0020] S13. Collect team data of entrepreneurial incubation projects and construct a team dataset D describing the characteristics of entrepreneurial incubation project teams t , including team-related indicators such as the number of team members, team stability, team professional ability matching, and team collaboration efficiency;

[0021] S14. Collect product data of entrepreneurial incubation projects and construct a product dataset D describing the product characteristics of entrepreneurial incubation projects p , including product-related indicators such as product innovation, product market demand adaptability, product iteration speed, and product technology maturity;

[0022] S15. Combine the financial data set, market data set, team data set and product data set constructed in steps S11 to S14 to construct a multi-dimensional data set:

[0023] D={D f ,D m ,D t ,D p}.

[0024] Optionally, S2 includes the following steps:

[0025] S21. Detecting records with obvious outliers in the multidimensional data set D, and filtering outliers based on statistical methods to obtain a cleaned multidimensional data set;

[0026] S22. Filling the missing data in the cleaned multidimensional data set to form a complete multidimensional data set;

[0027] S23. Normalize the filled complete multi-dimensional data set, unify the data scales of different dimensional features, and use the Min-Max normalization method to map the value range of numerical features to the interval [0,1];

[0028] S24. Integrate the normalized multidimensional data set into the preprocessed multidimensional data set D preprocessed .

[0029] Optionally, S3 includes the following steps:

[0030] S31. From the multidimensional dataset D preprocessed Extract categorical evaluation features and numerical evaluation features from the dataset and construct an evaluation feature set:

[0031] F={F c ,F n};

[0032] Among them, F c is the categorical evaluation feature set, F n is a set of numerical evaluation features;

[0033] S32. Quantify the impact of the evaluation feature set F of the entrepreneurial incubation project in the multi-stage decision-making process, and capture the global contribution of the evaluation features to the evaluation results by combining split nodes and weight calculation:

[0034]

[0035] Among them, I(F i ) represents the evaluation feature F i The importance value of T represents the set of all split nodes in the training data for the evaluation feature, G t (F i ) represents the evaluation feature F i Gain contribution at node t, W t represents the weight on node t, r and s are the number of categorical and numerical evaluation features respectively;

[0036] S33. According to the evaluation feature importance value I(F i ) and screening threshold θ, to screen out the optimal evaluation feature set F that meets the evaluation requirements selected :

[0037]

[0038] Among them, F selected It represents the selected optimal evaluation feature set, θ is the evaluation feature screening threshold, and α represents the dynamic weight coefficient, which is adjusted according to the stage of the entrepreneurial incubation project.

[0039] Optionally, S4 includes the following steps:

[0040] S41. Based on the optimal evaluation feature set F selected According to the start-up stage, growth stage and expansion stage of the entrepreneurial incubation project, a phased evaluation model M is constructed. stage :

[0041] M stage ={M1,M2,M3};

[0042] Among them, M1, M2, and M3 are stage-by-stage evaluation sub-models for the start-up stage, growth stage, and expansion stage, respectively. Each sub-model is based on the feature subset F of the corresponding stage. stage is the input feature set, defined as:

[0043] F stage ={F selected |Feature screening results with high stage correlation}.

[0044] S42. For each stage sub-model M k ,k∈{1,2,3}, according to the feature subset F stage The statistical characteristics of the model dynamically initialize the model parameters, including the learning rate η k 、Decision tree depth d k , number of iterations T k and the feature weight matrix W k :

[0045]

[0046]

[0047]

[0048] W k =Normalize(Corr(F stage ,y));

[0049] Among them, Var(F stage ) is the variance of the phase feature set, Corr(F stage ,y) represents the correlation matrix between the feature and the target variable y, W k It is the normalized feature weight matrix, which is used to dynamically adjust the model's attention to different features;

[0050] S43. Multi-objective optimization method is used to evaluate the staged model M stage Perform joint training while optimizing the evaluation accuracy, robustness, and computational efficiency of the staged evaluation model:

[0051]

[0052] in, represents the multi-objective loss function, represents the loss of evaluation accuracy, represents the stage-by-stage noise sensitivity loss, represents the resource load rate loss, α1, β, γ are the weight coefficients of multi-objective loss, which are set according to the actual needs of the entrepreneurial incubation project;

[0053] S44. Staged evaluation of the training completed sub-model M k Integrate and build a comprehensive initial assessment model M trained :

[0054]

[0055] Among them, ω k It is the integration weight of the staged sub-model, which is dynamically adjusted according to the evaluation accuracy and data coverage of each stage.

[0056] Optionally, the construction of the multi-stage success probability deviation loss includes evaluating the model M in stages stage The deviation between the predicted success probability and the actual success result is calculated for the start-up stage, growth stage, and expansion stage of each stage, and the weighted sum is obtained to obtain the multi-stage success probability deviation loss Used to measure the prediction accuracy of entrepreneurial incubation projects at different incubation stages:

[0057]

[0058] Where N is the total number of training samples, each sample corresponds to a real entrepreneurial incubation project, k is the stage index, and the value is 1 for the start-up stage, 2 for the growth stage, or 3 for the expansion stage. represents the predicted success probability of the i-th project in the k-th stage, represents the probability value corresponding to the actual success result or historical performance of the i-th project in the k-th stage, ω k It is the stage weight, which is used to highlight the relative importance of different stages in the evaluation;

[0059] Inject external and internal disturbance noise during model training to simulate policy fluctuations, market changes, and team instability, and calculate the difference in model output before and after noise to obtain the phased noise sensitivity loss Used to measure the evaluation robustness of the model in the face of real environment interference at different stages:

[0060]

[0061] Among them, ∈ represents the noise that simulates external or internal interference, including market environment fluctuations, team instability factors or product technology risks, It represents the probability of success of the model predicting the i-th project in the k-th stage after the introduction of noise ∈, represents the prediction success probability in the absence of noise, ν k is the noise sensitivity coefficient, which is dynamically set according to the susceptibility of the k-th stage entrepreneurial incubation project to external interference;

[0062] The phased noise sensitivity loss function measures the difference in the model prediction value before and after noise injection, examines the model's anti-interference ability at different incubation stages, and helps evaluate the model's robustness under real market or team fluctuations;

[0063] Build resource load rate loss based on the usage of system resources during training and inference The normalization process is used to calculate the time and resource consumption ratio of each staged sub-model during training and evaluation, which is used to balance the evaluation performance of the model and the actual deployment cost in the entrepreneurial incubation environment:

[0064]

[0065] Among them, τ k represents the computation time required for the k-th stage model to perform training or prediction, φ k represents the main system resource overhead occupied by the k-th stage model during training or prediction, τ max and φ max are the maximum acceptable computing time and maximum system resource overhead, C k represents the resource load coefficient of the k-th stage model, ρ k is the resource constraint coefficient. Different resource sensitivities can be set for different stages. Z represents the normalized constant.

[0066] Optionally, the S6 comprises the following steps:

[0067] S61. Calculate the influence of key evaluation features on the evaluation results through back propagation. By back propagating each layer of the optimized evaluation model, calculate the global gradient value of the input features to the output results, and identify the key evaluation feature set F with the top n feature importance rankings. key ;

[0068] S62. Based on the key evaluation feature set F key Perform disturbance simulation on the variation range of characteristic values ​​and calculate the risk sensitivity of risk points to the assessment results:

[0069]

[0070] Among them, R(F j ) represents the eigenvalue F j The risk sensitivity of j =∈ represents the eigenvalue F j The small disturbance introduced, and Respectively represent the evaluation results when disturbance is introduced and when no disturbance is introduced;

[0071] S63. Comprehensive key feature set F key and the corresponding risk sensitivity R(F j ) Construct the risk sensitivity analysis module R, the functions of the risk sensitivity analysis module include:

[0072] Risk point screening, based on sensitivity threshold θ r , filter out features with sensitivity exceeding the threshold as key risk points:

[0073] F risk ={F j ∣R(F j )≥θ r ,F j ∈F key};

[0074] Risk point ranking, according to risk sensitivity R (F j ) Sort the key risk points and give priority to marking the high-sensitivity risk points;

[0075] Dynamic risk warning, combined with real-time data of entrepreneurial incubation projects, monitors the dynamic changes of key risk points in real time, and triggers an alarm signal when the risk sensitivity exceeds the warning threshold.

[0076] The beneficial effects of the present invention are:

[0077] (1) The present invention constructs evaluation sub-models for the start-up stage, growth stage and expansion stage in stages, and combines them with a dynamic weight allocation mechanism to achieve accurate evaluation of different development stages of entrepreneurial incubation projects. The method of dynamically initializing model parameters based on characteristic statistical characteristics significantly improves the sensitivity and adaptability of the model to stage characteristics. In addition, through the integrated output of the staged evaluation sub-models, the evaluation results of different stages can be integrated to generate a more comprehensive evaluation report.

[0078] (2) The present invention introduces external and internal disturbance noise simulation technology. By simulating real scenarios of policy fluctuations, market changes and team instability, the difference in model output before and after noise is calculated, and a phased noise sensitivity loss function is constructed. This significantly improves the robustness of the evaluation model in complex and changing environments, and can effectively identify the performance differences of projects under risk scenarios, thereby providing more reliable decision support for incubators.

[0079] (3) The present invention constructs a risk sensitivity analysis module, which quantifies the global contribution of key assessment features to the assessment results by optimizing the back propagation calculation of the assessment model. At the same time, it identifies the key risk points that have the most significant impact on the assessment results through feature perturbation sensitivity analysis, and provides ranking and optimization suggestions. It not only provides a visual risk point analysis report, but also can dynamically track and warn of the changing trends of highly sensitive risk points. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0081] Figure 1 The present invention provides a flowchart of a method for evaluating the phased development of an entrepreneurial incubation project based on decision tree analysis. DETAILED DESCRIPTION

[0082] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0083] refer to Figure 1 , a phased development evaluation method for entrepreneurial incubation projects based on decision tree analysis, including the following steps:

[0084] S1. Collect multi-dimensional data sets related to entrepreneurial incubation projects, including financial data, market data, team data and product data;

[0085] S2. Clean the collected multidimensional data set, remove noise data, fill in missing data, perform normalization, unify the data scales of different dimensional features, and obtain the preprocessed multidimensional data set;

[0086] S3. extracting a variety of evaluation features from the preprocessed multidimensional data set, optimizing the extracted evaluation features, and forming a set of preferred evaluation features;

[0087] S4. construct a staged evaluation model using the evaluation feature set, and train the staged evaluation model to form an initial evaluation model;

[0088] S5. Use the validation data set to validate the trained initial evaluation model, evaluate the accuracy and robustness of the initial evaluation model, adjust the model parameters according to the validation results, optimize the feature interaction ability and the predictive ability of the staged development of the initial evaluation model, and obtain the optimized evaluation model;

[0089] S6. Construct a risk sensitivity analysis module based on the optimization evaluation model. By optimizing the back propagation calculation of the evaluation model, identify the key evaluation features that have the greatest impact on the evaluation results during the phased development of the entrepreneurial incubation project, conduct quantitative analysis on the key evaluation features, and generate sensitivity analysis results for key risk points;

[0090] S7. Input the real-time multi-dimensional data of the entrepreneurial incubation project to be evaluated into the risk sensitivity analysis module, calculate the phased development evaluation results of the entrepreneurial incubation project based on the optimization evaluation model, identify the key risk points in the development of the entrepreneurial incubation project according to the output of the risk sensitivity analysis module, and generate an evaluation report including the probability of success, potential risks and development bottlenecks.

[0091] In this implementation, S1 includes the following steps:

[0092] S11. Collect financial data of entrepreneurial incubation projects and construct a financial dataset D describing the financial characteristics of entrepreneurial incubation projects f , including financial indicators of total revenue, total expenditure, cash flow and capital utilization;

[0093] S12. Collect market data of entrepreneurial incubation projects and construct a market data set D that describes the market characteristics of entrepreneurial incubation projects m , including market-related indicators such as market coverage, customer conversion rate, market share and market growth rate;

[0094] S13. Collect team data of entrepreneurial incubation projects and construct a team dataset D describing the characteristics of entrepreneurial incubation project teams t , including team-related indicators such as the number of team members, team stability, team professional ability matching, and team collaboration efficiency;

[0095] S14. Collect product data of entrepreneurial incubation projects and construct a product dataset D describing the product characteristics of entrepreneurial incubation projects p , including product-related indicators such as product innovation, product market demand adaptability, product iteration speed, and product technology maturity;

[0096] S15. Combine the financial data set, market data set, team data set and product data set constructed in steps S11 to S14 to construct a multi-dimensional data set:

[0097] D={D f ,D m ,D t ,D p}.

[0098] In this implementation, S2 includes the following steps:

[0099] S21. Detecting records with obvious outliers in the multidimensional data set D, and filtering outliers based on statistical methods to obtain a cleaned multidimensional data set;

[0100] S22. Filling the missing data in the cleaned multidimensional data set to form a complete multidimensional data set;

[0101] S23. Normalize the filled complete multi-dimensional data set, unify the data scales of different dimensional features, and use the Min-Max normalization method to map the value range of numerical features to the interval [0,1];

[0102] S24. Integrate the normalized multidimensional data set into the preprocessed multidimensional data set D preprocessed .

[0103] In this implementation, S3 includes the following steps:

[0104] S31. From the multidimensional dataset D preprocessed Extract categorical evaluation features and numerical evaluation features from the dataset and construct an evaluation feature set:

[0105] F={F c ,F n};

[0106] Among them, F c is the categorical evaluation feature set, F n is a set of numerical evaluation features;

[0107] S32. Quantify the impact of the evaluation feature set F of the entrepreneurial incubation project in the multi-stage decision-making process, and capture the global contribution of the evaluation features to the evaluation results by combining split nodes and weight calculation:

[0108]

[0109] Among them, I(F i ) represents the evaluation feature F i The importance value of T represents the set of all split nodes in the training data for the evaluation feature, G t (F i ) represents the evaluation feature F i Gain contribution at node t, W t represents the weight on node t, r and s are the number of categorical and numerical evaluation features respectively;

[0110] S33. According to the evaluation feature importance value I(F i ) and screening threshold θ, to screen out the optimal evaluation feature set F that meets the evaluation requirements selected :

[0111]

[0112] Among them, F selected It represents the selected optimal evaluation feature set, θ is the evaluation feature screening threshold, and α represents the dynamic weight coefficient, which is adjusted according to the stage of the entrepreneurial incubation project.

[0113] In this implementation, S4 includes the following steps:

[0114] S41. Based on the optimal evaluation feature set F selected According to the start-up stage, growth stage and expansion stage of the entrepreneurial incubation project, a phased evaluation model M is constructed. stage :

[0115] M stage ={M1,M2,M3};

[0116] Among them, M1, M2, and M3 are stage-by-stage evaluation sub-models for the start-up stage, growth stage, and expansion stage, respectively. Each sub-model is based on the feature subset F of the corresponding stage. stage is the input feature set, defined as:

[0117] F stage ={F selected |Feature screening results with high stage correlation}.

[0118] S42. For each stage sub-model M k ,k∈{1,2,3}, according to the feature subset F stage The statistical characteristics of the model dynamically initialize the model parameters, including the learning rate η k 、Decision tree depth d k , number of iterations T k and the feature weight matrix W k :

[0119]

[0120]

[0121]

[0122] W k =Normalize(Corr(F stage ,y));

[0123] Among them, Var(F stage ) is the variance of the phase feature set, Corr(F stage ,y) represents the correlation matrix between the feature and the target variable y, W kIt is the normalized feature weight matrix, which is used to dynamically adjust the model's attention to different features;

[0124] S43. Multi-objective optimization method is used to evaluate the staged model M stage Perform joint training while optimizing the evaluation accuracy, robustness, and computational efficiency of the staged evaluation model:

[0125]

[0126] in, represents the multi-objective loss function, represents the loss of evaluation accuracy, represents the stage-by-stage noise sensitivity loss, represents the resource load rate loss, α1, β, γ are the weight coefficients of multi-objective loss, which are set according to the actual needs of the entrepreneurial incubation project;

[0127] S44. Staged evaluation of the training completed sub-model M k Integrate and build a comprehensive initial assessment model M trained :

[0128]

[0129] Among them, ω k It is the integration weight of the staged sub-model, which is dynamically adjusted according to the evaluation accuracy and data coverage of each stage.

[0130] In this embodiment, the construction of the multi-stage success probability deviation loss includes the following steps: stage The deviation between the predicted success probability and the actual success result is calculated for the start-up stage, growth stage, and expansion stage of each stage, and the weighted sum is obtained to obtain the multi-stage success probability deviation loss Used to measure the prediction accuracy of entrepreneurial incubation projects at different incubation stages:

[0131]

[0132] Where N is the total number of training samples, each sample corresponds to a real entrepreneurial incubation project, k is the stage index, and the value is 1 for the start-up stage, 2 for the growth stage, or 3 for the expansion stage. represents the predicted success probability of the i-th project in the k-th stage, represents the probability value corresponding to the actual success result or historical performance of the i-th project in the k-th stage, ω k It is the stage weight, which is used to highlight the relative importance of different stages in the evaluation;

[0133] Inject external and internal disturbance noise during model training to simulate policy fluctuations, market changes, and team instability, and calculate the difference in model output before and after noise to obtain the phased noise sensitivity loss Used to measure the evaluation robustness of the model in the face of real environment interference at different stages:

[0134]

[0135] Among them, ∈ represents the noise that simulates external or internal interference, including market environment fluctuations, team instability factors or product technology risks, It represents the probability of success of the model predicting the i-th project in the k-th stage after the introduction of noise ∈, represents the prediction success probability in the absence of noise, ν k is the noise sensitivity coefficient, which is dynamically set according to the susceptibility of the k-th stage entrepreneurial incubation project to external interference;

[0136] The phased noise sensitivity loss function measures the difference in the model prediction value before and after noise injection, examines the model's anti-interference ability at different incubation stages, and helps evaluate the model's robustness under real market or team fluctuations;

[0137] Build resource load rate loss based on the usage of system resources during training and inference The normalization process is used to calculate the time and resource consumption ratio of each staged sub-model during training and evaluation, which is used to balance the evaluation performance of the model and the actual deployment cost in the entrepreneurial incubation environment:

[0138]

[0139] Among them, τ k represents the computation time required for the k-th stage model to perform training or prediction, φ k represents the main system resource overhead occupied by the k-th stage model during training or prediction, τ max and φ max are the maximum acceptable computing time and maximum system resource overhead, C k represents the resource load coefficient of the k-th stage model, ρ k is the resource constraint coefficient. Different resource sensitivities can be set for different stages. Z represents the normalized constant.

[0140] In this implementation, S6 includes the following steps:

[0141] S61. Calculate the influence of key evaluation features on the evaluation results through back propagation. By back propagating each layer of the optimized evaluation model, calculate the global gradient value of the input features to the output results, and identify the key evaluation feature set F with the top n feature importance rankings. key ;

[0142] S62. Based on the key evaluation feature set F key Perform disturbance simulation on the variation range of characteristic values ​​and calculate the risk sensitivity of risk points to the assessment results:

[0143]

[0144] Among them, R(F j ) represents the eigenvalue F j The risk sensitivity of j =∈ represents the eigenvalue F j The small disturbance introduced, and Respectively represent the evaluation results when disturbance is introduced and when no disturbance is introduced;

[0145] S63. Comprehensive key feature set F key and the corresponding risk sensitivity R(F j ) Construct the risk sensitivity analysis module R, the functions of the risk sensitivity analysis module include:

[0146] Risk point screening, based on sensitivity threshold θ r , filter out features with sensitivity exceeding the threshold as key risk points:

[0147] F risk ={F j ∣R(F j )≥θ r ,F j ∈F key};

[0148] Risk point ranking, according to risk sensitivity R (F j ) Sort the key risk points and give priority to marking the high-sensitivity risk points;

[0149] Dynamic risk warning, combined with real-time data of entrepreneurial incubation projects, monitors the dynamic changes of key risk points in real time, and triggers an alarm signal when the risk sensitivity exceeds the warning threshold.

[0150] Embodiment 1:

[0151] Embodiment In March 2024, a national-level entrepreneurship incubation center found that the quarterly development report submitted by the incubated enterprise "Project A" showed that the proportion of its R&D expenditure to revenue fluctuated greatly, and the market expansion goals were not completed as planned. In order to evaluate the actual status of the current development stage of "Project A" and identify potential risks, the incubation center decided to apply the method of the present invention to conduct a phased development evaluation on it. The following is a detailed record of the complete evaluation process:

[0152] On March 12, 2024, the Incubation Center extracted the following multi-dimensional data from the financial, market, team and product management systems of "Project A":

[0153] Financial data: Revenue of 1.2 million yuan in the first quarter of 2024, R&D expenditure accounts for 50%, and cash flow fluctuation range is ±20%;

[0154] Market data: customer conversion rate 20%, market share 3%, market growth rate 5%;

[0155] Team data: Team size: 20 people, 4 people have left in the past 3 months, and the team collaboration efficiency score is 60%;

[0156] Product data: The version iteration cycle of the core product "Smart Farmland Management Platform" is 90 days, and the market demand adaptability is 80%.

[0157] During the data cleaning process, it was found that some records of team collaboration efficiency scores were missing. The data was repaired by filling the mean, and all values ​​were mapped to the interval [0,1] through normalization.

[0158] On March 13, 2024, the system extracted 30 features including capital utilization, market growth rate, team stability, and product innovation from the preprocessed data set. Through feature importance calculation, it identified 10 key features that have the greatest impact on the evaluation results, such as capital utilization (weight 0.85), team collaboration efficiency (weight 0.75), and market fit (weight 0.65).

[0159] On March 14, 2024, the Incubation Center constructed a training set by calling historical data, including 150 similar projects incubated during 2018-2024. The training data contained the true success probability labels and their corresponding feature values ​​at each stage. The feature weights of the start-up model were dynamically adjusted, and the capital utilization rate was given the highest priority (weight 0.9). The training process used a multi-objective optimization algorithm to generate a comprehensive initial evaluation model.

[0160] On March 15, 2024, the optimization evaluation model conducted real-time data input and prediction analysis on "Project A":

[0161] The predicted success probability in the start-up phase is 72%;

[0162] The growth stage predicts a success probability of 64%;

[0163] The expansion phase predicts a 55% probability of success.

[0164] The system further identified the following risk points:

[0165] 1. Abnormal fund utilization: Fund expenditure in the past three months has not reached the established plan, and the fund interruption risk score is 0.85.

[0166] 2. Low team stability: The departure of four key technical personnel has led to a decline in the team’s core capabilities, with a risk score of 0.78.

[0167] 3. Market target deviation: The market expansion plan was not completed, resulting in a decrease in the probability of success in the expansion phase, with a risk score of 0.71.

[0168] The system generated a detailed sensitivity analysis report, recording the impact value of each key risk point and the recommended optimization plan. For example, in response to the risk of funding shortage, the system suggested that the incubation center provide an additional 500,000 yuan of funding support for "Project A" and introduce a fund use monitoring module.

[0169] On March 16, 2024, based on the analysis report provided by the system, the incubation center decided to add 500,000 yuan in financial support in the next two months to give priority to solving the risk of funding rupture. At the same time, it organized a team of marketing experts to assist the company in adjusting its market expansion strategy and recommended one high-end technical talent to fill the vacancy of the core team members.

[0170] This example fully verifies the significant advantages of the present invention in terms of assessment accuracy, risk identification capability, and resource optimization support through a real entrepreneurial incubation project case. The incubation center effectively identified and solved the core problems of "Project A" through the method of the present invention, significantly improving its success rate of incubation.

[0171] The present invention constructs evaluation sub-models for the start-up stage, growth stage and expansion stage in stages, and combines them with a dynamic weight allocation mechanism to achieve accurate evaluation of different development stages of entrepreneurial incubation projects. The method of dynamically initializing model parameters based on characteristic statistical characteristics significantly improves the sensitivity and adaptability of the model to stage characteristics. In addition, through the integrated output of the staged evaluation sub-models, the evaluation results of different stages can be integrated to generate a more comprehensive evaluation report.

[0172] The present invention introduces external and internal disturbance noise simulation technology. By simulating real scenarios of policy fluctuations, market changes and team instability, the difference in model output before and after noise is calculated, and a phased noise sensitivity loss function is constructed. This significantly improves the robustness of the evaluation model in complex and changing environments, and can effectively identify performance differences of projects under risk scenarios, thereby providing more reliable decision-making support for incubators.

[0173] The present invention constructs a risk sensitivity analysis module, which quantifies the global contribution of key evaluation features to the evaluation results by optimizing the back propagation calculation of the evaluation model. At the same time, it identifies the key risk points that have the most significant impact on the evaluation results through feature perturbation sensitivity analysis, and provides sorting and optimization suggestions. It not only provides a visual risk point analysis report, but also can dynamically track and warn of the changing trends of highly sensitive risk points.

[0174] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for evaluating the phased development of entrepreneurial incubation projects based on decision tree analysis, characterized in that: The steps include: S1. Collect multi-dimensional data sets related to entrepreneurial incubation projects, including financial data, market data, team data and product data; S2. Clean the collected multidimensional data set, remove noise data, fill in missing data, perform normalization, unify the data scales of different dimensional features, and obtain the preprocessed multidimensional data set; S3. extracting a variety of evaluation features from the preprocessed multidimensional data set, optimizing the extracted evaluation features, and forming a set of preferred evaluation features; S4. construct a staged evaluation model using the evaluation feature set, and train the staged evaluation model to form an initial evaluation model; S5. Use the validation data set to validate the trained initial evaluation model, evaluate the accuracy and robustness of the initial evaluation model, adjust the model parameters according to the validation results, optimize the feature interaction ability and the predictive ability of the staged development of the initial evaluation model, and obtain the optimized evaluation model; S6. Construct a risk sensitivity analysis module based on the optimization evaluation model. By optimizing the back propagation calculation of the evaluation model, identify the key evaluation features that have the greatest impact on the evaluation results during the phased development of the entrepreneurial incubation project, conduct quantitative analysis on the key evaluation features, and generate sensitivity analysis results for key risk points; S7. Input the real-time multi-dimensional data of the entrepreneurial incubation project to be evaluated into the risk sensitivity analysis module, calculate the phased development evaluation results of the entrepreneurial incubation project based on the optimization evaluation model, identify the key risk points in the development of the entrepreneurial incubation project according to the output of the risk sensitivity analysis module, and generate an evaluation report including the probability of success, potential risks and development bottlenecks.

2. According to claim 1, a method for evaluating the phased development of an entrepreneurial incubation project based on decision tree analysis is characterized in that: The S1 comprises the following steps: S11. Collect financial data of entrepreneurial incubation projects and construct a financial dataset D describing the financial characteristics of entrepreneurial incubation projects f , including financial indicators of total revenue, total expenditure, cash flow and capital utilization; S12. Collect market data of entrepreneurial incubation projects and construct a market data set D that describes the market characteristics of entrepreneurial incubation projects m , including market-related indicators such as market coverage, customer conversion rate, market share and market growth rate; S13. Collect team data of entrepreneurial incubation projects and construct a team dataset D describing the characteristics of entrepreneurial incubation project teams t , including team-related indicators such as the number of team members, team stability, team professional ability matching, and team collaboration efficiency; S14. Collect product data of entrepreneurial incubation projects and construct a product dataset D describing the product characteristics of entrepreneurial incubation projects p , including product-related indicators such as product innovation, product market demand adaptability, product iteration speed, and product technology maturity; S15. Combine the financial data set, market data set, team data set and product data set constructed in steps S11 to S14 to construct a multi-dimensional data set: D={D f ,D m ,D t ,D p }。 3. According to claim 1, a method for evaluating the phased development of an entrepreneurial incubation project based on decision tree analysis is characterized in that: The S2 comprises the following steps: S21. Detecting records with obvious outliers in the multidimensional data set D, and filtering outliers based on statistical methods to obtain a cleaned multidimensional data set; S22. Filling the missing data in the cleaned multidimensional data set to form a complete multidimensional data set; S23. Normalize the filled complete multi-dimensional data set, unify the data scales of different dimensional features, and use the Min-Max normalization method to map the value range of numerical features to the interval [0,1]; S24. Integrate the normalized multidimensional data set into the preprocessed multidimensional data set D preprocessed .

4. According to claim 1, a method for evaluating the phased development of an entrepreneurial incubation project based on decision tree analysis is characterized in that: The S3 comprises the following steps: S31. From the multidimensional dataset D preprocessed Extract categorical evaluation features and numerical evaluation features from the dataset and construct an evaluation feature set: F={F c ,F n }; Among them, F c is the categorical evaluation feature set, F n is a set of numerical evaluation features; S32. Quantify the impact of the evaluation feature set F of the entrepreneurial incubation project in the multi-stage decision-making process, and capture the global contribution of the evaluation features to the evaluation results by combining split nodes and weight calculation: Among them, I(F i ) represents the evaluation feature F i The importance value of T represents the set of all split nodes in the training data for the evaluation feature, G t (F i ) represents the evaluation feature F i Gain contribution at node t, W t represents the weight on node t, r and s are the number of categorical and numerical evaluation features respectively; S33. According to the evaluation feature importance value I(F i ) and screening threshold θ, to screen out the optimal evaluation feature set F that meets the evaluation requirements selected : Among them, F selected It represents the selected optimal evaluation feature set, θ is the evaluation feature screening threshold, and α represents the dynamic weight coefficient, which is adjusted according to the stage of the entrepreneurial incubation project.

5. According to claim 1, a method for evaluating the phased development of an entrepreneurial incubation project based on decision tree analysis is characterized in that: The S4 comprises the following steps: S41. Based on the optimal evaluation feature set F selected According to the start-up stage, growth stage and expansion stage of the entrepreneurial incubation project, a phased evaluation model M is constructed. stage : <h2 style=";text-align:left;direction:ltr">M<h2 style=";text-align:left;direction:ltr"> stage <h2 style=";text-align:left;direction:ltr"> (M1,M2,M3) Among them, M1, M2, and M3 are stage-by-stage evaluation sub-models for the start-up stage, growth stage, and expansion stage, respectively. Each sub-model is based on the feature subset F of the corresponding stage. stage is the input feature set, defined as: F stage ={F selected |Feature screening results with high stage correlation}. S42. For each stage sub-model M k ,k∈{1,2,3}, according to the feature subset F stage The statistical characteristics of the model dynamically initialize the model parameters, including the learning rate η k 、Decision tree depth d k , number of iterations T k and the feature weight matrix W k : W k =Normalize(Corr(F stage ,y)); Among them, Var(F stage ) is the variance of the phase feature set, Corr(F stage ,y) represents the correlation matrix between the feature and the target variable y, W k It is the normalized feature weight matrix, which is used to dynamically adjust the model's attention to different features; S43. Multi-objective optimization method is used to evaluate the staged model M stage Perform joint training while optimizing the evaluation accuracy, robustness, and computational efficiency of the staged evaluation model: in, represents the multi-objective loss function, represents the loss of evaluation accuracy, represents the stage-by-stage noise sensitivity loss, represents the resource load rate loss, α1, β, γ are the weight coefficients of multi-objective loss, which are set according to the actual needs of the entrepreneurial incubation project; S44. Staged evaluation of the training completed sub-model M k Integrate and build a comprehensive initial assessment model M trained : Among them, ω k It is the integration weight of the staged sub-model, which is dynamically adjusted according to the evaluation accuracy and data coverage of each stage.

6. The method for evaluating the phased development of an entrepreneurial incubation project based on decision tree analysis according to claim 5 is characterized in that: The construction of the multi-stage success probability deviation loss includes the following steps: stage The deviation between the predicted success probability and the actual success result is calculated for the start-up stage, growth stage, and expansion stage of each stage, and the weighted sum is obtained to obtain the multi-stage success probability deviation loss L accuracy , used to measure the prediction accuracy of entrepreneurial incubation projects at different incubation stages: Where N is the total number of training samples, each sample corresponds to a real entrepreneurial incubation project, k is the stage index, and the value is 1 for the start-up stage, 2 for the growth stage, or 3 for the expansion stage. represents the predicted success probability of the i-th project in the k-th stage, represents the probability value corresponding to the actual success result or historical performance of the i-th project in the k-th stage, ω k It is the stage weight, which is used to highlight the relative importance of different stages in the evaluation; Inject external and internal disturbance noise during model training to simulate policy fluctuations, market changes, and team instability, and calculate the difference in model output before and after noise to obtain the phased noise sensitivity loss Used to measure the evaluation robustness of the model in the face of real environment interference at different stages: Among them, ∈ represents the noise that simulates external or internal interference, including market environment fluctuations, team instability factors or product technology risks, It represents the probability of success of the model predicting the i-th project in the k-th stage after the introduction of noise ∈, represents the prediction success probability in the absence of noise, ν k is the noise sensitivity coefficient, which is dynamically set according to the susceptibility of the k-th stage entrepreneurial incubation project to external interference; The phased noise sensitivity loss function measures the difference in the model prediction value before and after noise injection, examines the model's anti-interference ability at different incubation stages, and helps evaluate the model's robustness under real market or team fluctuations; Build resource load rate loss based on the usage of system resources during training and inference The normalization process is used to calculate the time and resource consumption ratio of each staged sub-model during training and evaluation, which is used to balance the evaluation performance of the model and the actual deployment cost in the entrepreneurial incubation environment: Among them, τ k represents the computation time required for the k-th stage model to perform training or prediction, φ k represents the main system resource overhead occupied by the k-th stage model during training or prediction, τ max and φ max are the maximum acceptable computing time and maximum system resource overhead, C k represents the resource load coefficient of the k-th stage model, ρ k is the resource constraint coefficient. Different resource sensitivities can be set for different stages. Z represents the normalized constant.

7. The method for evaluating the phased development of an entrepreneurial incubation project based on decision tree analysis according to claim 1 is characterized in that: The S6 comprises the following steps: S61. Calculate the influence of key evaluation features on the evaluation results through back propagation. By back propagating each layer of the optimized evaluation model, calculate the global gradient value of the input features to the output results, and identify the key evaluation feature set F with the top n feature importance rankings. key ; S62. Based on the key evaluation feature set F key Perform disturbance simulation on the variation range of characteristic values ​​and calculate the risk sensitivity of risk points to the assessment results: Among them, R(F j ) represents the eigenvalue F j The risk sensitivity of j =∈ represents the eigenvalue F j The small disturbance introduced, and Respectively represent the evaluation results when disturbance is introduced and when no disturbance is introduced; S63. Comprehensive key feature set F key and the corresponding risk sensitivity R(F j ) Construct the risk sensitivity analysis module R, the functions of the risk sensitivity analysis module include: Risk point screening, based on sensitivity threshold θ r , filter out features with sensitivity exceeding the threshold as key risk points: F risk ={F j ∣R(F j )≥θ r ,F j ∈F key }; Risk point ranking, according to risk sensitivity R (F j ) Sort the key risk points and give priority to marking the high-sensitivity risk points; Dynamic risk warning, combined with real-time data of entrepreneurial incubation projects, monitors the dynamic changes of key risk points in real time, and triggers an alarm signal when the risk sensitivity exceeds the warning threshold.

Images