CN111738870B - Method and platform for identifying insurance risk of engineering performance guarantee based on characteristic engineering - Google Patents

Method and platform for identifying insurance risk of engineering performance guarantee based on characteristic engineering Download PDF

Info

Publication number
CN111738870B
CN111738870B CN202010739603.9A CN202010739603A CN111738870B CN 111738870 B CN111738870 B CN 111738870B CN 202010739603 A CN202010739603 A CN 202010739603A CN 111738870 B CN111738870 B CN 111738870B
Authority
CN
China
Prior art keywords
feature
engineering
data
features
model
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202010739603.9A
Other languages
Chinese (zh)
Other versions
CN111738870A (en
Inventor
曾雪强
谢仑辰
徐学武
史清江
陈海军
化允
陈华龙
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Gongbao Technology Zhejiang Co ltd
Original Assignee
Gongbao Technology Zhejiang Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Gongbao Technology Zhejiang Co ltd filed Critical Gongbao Technology Zhejiang Co ltd
Priority to CN202010739603.9A priority Critical patent/CN111738870B/en
Publication of CN111738870A publication Critical patent/CN111738870A/en
Application granted granted Critical
Publication of CN111738870B publication Critical patent/CN111738870B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/08—Insurance
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/067—Enterprise or organisation modelling

Landscapes

  • Business, Economics & Management (AREA)
  • Engineering & Computer Science (AREA)
  • Strategic Management (AREA)
  • Economics (AREA)
  • Human Resources & Organizations (AREA)
  • General Business, Economics & Management (AREA)
  • Marketing (AREA)
  • Development Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Physics & Mathematics (AREA)
  • Finance (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Accounting & Taxation (AREA)
  • Educational Administration (AREA)
  • Technology Law (AREA)
  • Game Theory and Decision Science (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Financial Or Insurance-Related Operations Such As Payment And Settlement (AREA)

Abstract

The invention discloses a method and a platform for identifying insurance risk of engineering performance guarantee based on characteristic engineering, which comprises the following steps of firstly, carrying out preprocessing operation on engineering service data, and constructing an initial training data set according to the preprocessed data; then, according to the initial training data set, utilizing an XGboost model to train to obtain a benchmark risk evaluation model; secondly, performing feature screening by utilizing a maximum mutual information feature selection strategy and a benchmark risk evaluation model aiming at the initial training data set to obtain a screened training data set, and training by using an XGboost model to obtain a final risk evaluation model; and finally, performing risk assessment on the item to be assessed by using the obtained risk assessment model. The method can find out key characteristics from a large amount of redundant engineering project data, and reduces the complexity of the model while ensuring the predictive performance of the model.

Description

Method and platform for identifying insurance risk of engineering performance guarantee based on characteristic engineering
Technical Field
The invention relates to the technical field of engineering insurance and machine learning, in particular to a method and a platform for identifying risk of engineering insurance for ensuring performance based on feature engineering.
Background
The construction process and the construction flow of the construction project are complex, the number of project participants is large, the project period is long, the related area is wide, and the default of a construction unit can cause loss in various aspects, so that the introduction of a wind control mechanism for ensuring insurance in the construction project is particularly important, the cash guarantee fund pressure can be effectively released by a construction enterprise, and the enterprise burden is relieved. For the insurance industry, the main difficult problem for developing construction engineering insurance assurance is data and wind control, and the lack of professional knowledge and technology of construction engineering projects for insurance companies leads to difficult assessment of risks of policemen, insurance projects and insureds. The non-financing type guarantees that the insurance approval speed is required to be high, and the insurance applicant, the engineering project and the insured cannot be comprehensively examined.
Risk factors causing the engineering default have the characteristics of diversity, universality, objectivity, contingency and the like, so that the number of risk factors for performing is large, and strong relevance exists among the risk factors. The current engineering insurance mainly uses manpower judgment, is long in time consumption, does not utilize extensive project data information, and is the defect of the current risk judgment method. The algorithm model of the invention utilizes a large amount of data information and an intelligent algorithm model to integrate and analyze risk factors of the policyholder, the engineering project and the insured, thereby really achieving the purpose of quickly identifying the default risk of the construction project and assisting the insurance company to reduce the underwriting risk.
Disclosure of Invention
The invention aims to provide a method and a platform for identifying the risk of insurance for ensuring the performance of an engineering based on characteristic engineering, aiming at the defects of the prior art.
The purpose of the invention is realized by the following technical scheme: a method for identifying the risk of insurance for ensuring the performance of an engineering based on characteristic engineering comprises the following steps:
s1: carrying out preprocessing operation on the engineering project information data to obtain engineering service data, and constructing an initial training data set according to the engineering service data;
s2: according to the initial training data set, a standard risk assessment model is obtained by utilizing XGboost model training, and the discrimination accuracy of the standard risk assessment model is recorded
Figure 890972DEST_PATH_IMAGE001
；
S3: and (3) performing feature screening by using feature engineering aiming at an initial training data set, wherein the feature engineering is a maximum correlation minimum redundancy combined maximum mutual information coefficient feature selection strategy, is recorded as MR-MIC, and is combined with a reference risk assessment model and the judgment accuracy rate thereof
Figure 321953DEST_PATH_IMAGE002
Obtaining a screened training data set; the method specifically comprises the following steps: firstly, calculating the maximum mutual information coefficient of each pair of characteristics and each characteristic and the corresponding class label in engineering service data, then constructing a characteristic index set, and recording the judgment accuracy rate of each characteristic index set
Figure 49517DEST_PATH_IMAGE003
Selecting the feature index set with the highest accuracy, and recording the highest discrimination accuracy
Figure 616765DEST_PATH_IMAGE004
Accuracy of discrimination from reference risk assessment model
Figure 508497DEST_PATH_IMAGE002
For comparison, if
Figure 36562DEST_PATH_IMAGE005
Then determining the selected feature index set as the finally selected feature index set, if so
Figure 671942DEST_PATH_IMAGE006
Then, the feature index set is sorted and traversed from large to small according to the feature number in the feature index set, and a feature index set is found, and the judgment accuracy rate is high
Figure 234642DEST_PATH_IMAGE003
Greater than the threshold of accuracy, the threshold of accuracy discriminates the accuracy according to
Figure 297276DEST_PATH_IMAGE002
And the required precision is selected, and the screened characteristic quantity is larger than the characteristic quantity threshold; performing feature screening based on the found feature index set to obtain a screened training data set;
s4: aiming at the screened training data set, using an XGboost model to train to obtain a final risk assessment model;
s5: after the engineering business data is obtained by the preprocessing operation in the step S1 on the information data of the engineering project to be evaluated, the MR-MIC feature screening in the step S3 is used, and then the engineering business data after preprocessing and feature screening is input to the final risk evaluation model obtained in the step S4, so as to obtain a risk evaluation result of the project to be evaluated.
Further, the preprocessing operation in step S1 specifically includes:
and carrying out one-hot coding processing on the class characteristics described in the form of characters in the engineering service data to obtain discrete numerical characteristics, and meanwhile, filling missing values in the characteristics described in the form of numerical values in the engineering service data by using a median filling method to finish data preprocessing.
Further, the feature screening policy in step S3 specifically includes:
s31: setting a mesh partition size parameterBProduce a satisfactionm*n<BVarious kinds of (A), (B), (Cm,n) A combination of positive integers of (a) is,mandnvalues for grid horizontal and vertical division;
s32: for each pair of characteristics in engineering service dataXAndYgo through each group (m,n) Will beXIs divided evenly intomShare and find the feature by using dynamic programmingXAndYfeatures with maximum mutual informationYIs then fixed to the featureYUsing dynamic programming to find the featuresXAndYfeatures with maximum mutual informationXIs divided, then, the feature is fixedXUsing dynamic programming to find the featuresXAndYfeatures with maximum mutual informationYAnd finally outputting each group of (m,n) Corresponding maximum mutual information valueI mn (X,Y)；
S33: each pair is calculated according to the following formulaXAndYmaximum mutual information coefficient of
Figure 171691DEST_PATH_IMAGE007
：
Figure 751708DEST_PATH_IMAGE008
Method for calculating maximum mutual information coefficient between each feature and corresponding class label in engineering service data and each pair of featuresXAndYthe maximum mutual information coefficient calculation methods are consistent;
s34: constructing feature index setsS 1：
Figure 27969DEST_PATH_IMAGE009
Wherein
Figure 261504DEST_PATH_IMAGE010
For the first in engineering business datakThe characteristics of the device are as follows,cis a category label;
Figure 764160DEST_PATH_IMAGE011
for the features calculated according to step S32 and step S33
Figure 475765DEST_PATH_IMAGE010
And its corresponding category labelcMaximum mutual information coefficient therebetween;
s35: generating the remaining feature index set by the following formula
Figure 872111DEST_PATH_IMAGE012
：
Figure 417493DEST_PATH_IMAGE013
WhereinTRepresenting the total number of features in the engineering business data;
Figure 266500DEST_PATH_IMAGE014
indexing sets for featuresS t Is indexed byiIs characterized in that it is a mixture of two or more of the above-mentioned components,
Figure 47374DEST_PATH_IMAGE015
indexing sets for unselected features
Figure 173593DEST_PATH_IMAGE016
The middle index isjThe features of (1);
s36: indexing each feature into a setS t Inputting the corresponding data set into the XGboost model, and recording the discrimination accuracy
Figure 748931DEST_PATH_IMAGE003
And selecting the feature index set with the highest accuracy
Figure 85234DEST_PATH_IMAGE017
Simultaneously recording the highest discrimination accuracy
Figure 545166DEST_PATH_IMAGE018
；
S37: will be provided with
Figure 650525DEST_PATH_IMAGE018
Determination accuracy of the reference risk assessment model in step S2
Figure 662343DEST_PATH_IMAGE019
Make a comparison if
Figure 92800DEST_PATH_IMAGE020
Then determine
Figure 215477DEST_PATH_IMAGE017
For the finally selected feature index set, if
Figure 440922DEST_PATH_IMAGE021
Then go from big to smalltFind onetThe accuracy of the discrimination
Figure 233429DEST_PATH_IMAGE003
Greater than a threshold of accuracy, i.e.
Figure 544324DEST_PATH_IMAGE022
And the number of features to be screened out is greater than the threshold number of features, i.e. the number of features to be screened out
Figure 470692DEST_PATH_IMAGE023
And determineS t As a final selected feature index set, wherein,aandbis a parameter set according to requirements;
s38: and performing feature screening based on the finally selected feature index set to obtain a screened training data set.
A project performance guarantee insurance risk identification platform based on feature engineering comprises a data input module, a data processing module, a feature calculation and screening module, a model training module and a risk assessment module:
the data input module is used for receiving engineering project information data needing risk identification, and the data input module comprises engineering project information data input for model training or engineering project information data to be evaluated;
the data processing module is used for executing preprocessing operation on the engineering project information data to obtain engineering service data, and generating an initial training data set or preprocessing the engineering project information data to be evaluated;
the characteristic calculation and screening module is used for carrying out characteristic screening on the initial training data set processed by the data processing module by utilizing characteristic engineering, the characteristic engineering is a maximum correlation minimum redundancy combined maximum mutual information coefficient characteristic selection strategy, which is recorded as MR-MIC, and a reference risk discrimination model obtained by combining the model training module and the discrimination accuracy rate thereof
Figure 160430DEST_PATH_IMAGE001
And (3) carrying out feature screening to obtain a screened training data set, which specifically comprises the following steps: firstly, calculating the maximum mutual information coefficient of each pair of characteristics and each characteristic and the corresponding class label in engineering service data, then constructing a characteristic index set, and recording the judgment accuracy rate of each characteristic index set
Figure 514051DEST_PATH_IMAGE003
Selecting the feature index set with the highest accuracy, and recording the highest discrimination accuracy
Figure 312243DEST_PATH_IMAGE004
Accuracy of discrimination from reference risk assessment model
Figure 386509DEST_PATH_IMAGE024
For comparison, if
Figure 586547DEST_PATH_IMAGE025
Then determining the selected feature index set as the finally selected feature index set, if so
Figure 111069DEST_PATH_IMAGE006
Then, the feature index set is sorted and traversed from large to small according to the feature number in the feature index set, and a feature index set is found, and the judgment accuracy rate is high
Figure 740764DEST_PATH_IMAGE003
Greater than the threshold of accuracy, the threshold of accuracy discriminates the accuracy according to
Figure 274514DEST_PATH_IMAGE001
And the required precision is selected, and the screened characteristic quantity is larger than the characteristic quantity threshold; performing feature screening based on the found feature index set to obtain a screened training data set;
the model training module is used for training the initial training data set processed by the data processing module by using the XGboost model to obtain a reference risk discrimination model and recording the discrimination accuracy of the reference risk discrimination model
Figure 63478DEST_PATH_IMAGE024
(ii) a Or training the screened training data set generated by the feature calculation and screening module by using an XGboost model to obtain a final risk discrimination model;
and the risk evaluation module is used for giving a risk judgment result of the information data of the engineering project to be evaluated, which is input by the data input module, according to the final risk evaluation model.
Furthermore, the data input module receives data input in a unified mode from the outside and stores the data in a database.
Further, the data processing module comprises a character characteristic processing module and a numerical characteristic processing module;
the character feature processing module is used for carrying out one-hot coding processing on the class features described in the form of characters in the engineering service data to obtain discrete numerical features;
and the numerical value characteristic processing module is used for filling missing values by using a median filling method aiming at the characteristics described in a numerical value form in the engineering service data.
Further, the feature calculating and screening module comprises a maximum mutual information coefficient calculating module, a feature index set generating module and a feature screening module;
the maximum mutual information coefficient calculation module is used for calculating each pair of characteristics in the engineering service data obtained by the data processing moduleXAndYor the maximum mutual information coefficient between each feature and its corresponding class label; the method comprises the following specific steps:
(1) setting a mesh partition size parameterBProduce a satisfactionm*n<BVarious kinds of (A), (B), (Cm,n) A combination of positive integers of (a) is,mandnvalues for grid horizontal and vertical division;
(2) for each pair of characteristics in engineering service dataXAndYgo through each group (m,n) Will beXIs divided evenly intomShare and find the feature by using dynamic programmingXAndYfeatures with maximum mutual informationYIs then fixed to the featureYUsing dynamic programming to find the featuresXAndYfeatures with maximum mutual informationXIs divided, then, the feature is fixedXUsing dynamic programming to find the featuresXAndYfeatures with maximum mutual informationYAnd finally outputting each group of (m,n) Corresponding maximum mutual information valueI mn (X,Y)；
(3) Each pair is calculated according to the following formulaXAndYmaximum mutual information coefficient of
Figure 634268DEST_PATH_IMAGE007
：
Figure 141473DEST_PATH_IMAGE026
Method for calculating maximum mutual information coefficient between each feature and corresponding class label in engineering service data and each pair of featuresXAndYthe maximum mutual information coefficient calculation methods are consistent;
the feature index set generation module is configured to perform feature screening on the data preprocessed by the data processing module by using an MR-MIC feature selection policy according to the maximum mutual information coefficients between each pair of features calculated by the maximum mutual information coefficient module and between each feature and the corresponding category label thereof, and generate all feature index sets, specifically as follows:
(1) constructing feature index setsS 1：
Figure 88700DEST_PATH_IMAGE027
Wherein
Figure 732171DEST_PATH_IMAGE010
For the first in engineering business datakThe characteristics of the device are as follows,cis a category label;
Figure 598496DEST_PATH_IMAGE011
for features obtained from the maximum mutual information coefficient calculation module
Figure 327418DEST_PATH_IMAGE010
And its corresponding category labelcMaximum of betweenA mutual information coefficient;
(2) generating the remaining feature index set by the following formula
Figure 78336DEST_PATH_IMAGE012
：
Figure 841893DEST_PATH_IMAGE028
WhereinTRepresenting the total number of features in the engineering business data;
Figure 613540DEST_PATH_IMAGE014
indexing sets for featuresS t Is indexed byiIs characterized in that it is a mixture of two or more of the above-mentioned components,
Figure 233352DEST_PATH_IMAGE015
indexing sets for unselected features
Figure 381437DEST_PATH_IMAGE029
The middle index isjThe features of (1);
the feature screening module is used for selecting the feature index set with the highest accuracy value from all the feature index sets obtained by the feature index set generation module
Figure 999500DEST_PATH_IMAGE017
Simultaneously recording the highest discrimination accuracy
Figure 82994DEST_PATH_IMAGE018
Accuracy of discrimination from reference risk assessment model
Figure 786507DEST_PATH_IMAGE019
For comparison, if
Figure 738283DEST_PATH_IMAGE030
Then determine
Figure 351798DEST_PATH_IMAGE031
For the finally selected feature index set, if
Figure 730827DEST_PATH_IMAGE032
Then go from big to smalltFind onetThe accuracy of the discrimination
Figure 62582DEST_PATH_IMAGE003
Greater than a threshold of accuracy, i.e.
Figure 552469DEST_PATH_IMAGE033
And the number of features to be screened out is greater than the threshold number of features, i.e. the number of features to be screened out
Figure 145125DEST_PATH_IMAGE023
And determineS t As a final selected feature index set, wherein,aandband performing feature screening on the parameters set according to requirements and based on the finally selected feature index set to obtain a screened training data set.
The invention has the beneficial effects that: the method utilizes an MR-MIC characteristic selection strategy, can find out the characteristics most relevant to the class labels from a large amount of engineering project data, and simultaneously ensures that the redundancy degree between the selected characteristics is lower, thereby reducing the complexity of the model while ensuring the predictive performance of the model. The XGboost algorithm is adopted to construct the model, so that the result accuracy of the proposed risk identification method is ensured.
Drawings
FIG. 1 is a flow chart of a method for identifying insurance risk of project performance guarantee based on feature engineering provided by the present invention;
FIG. 2 is a schematic structural diagram of an engineering performance guarantee insurance risk identification platform based on feature engineering according to the present invention;
FIG. 3 is a diagram of a feature of the area of insurance for ensuring the performance of an engineering project.
Detailed Description
The invention will be further described with reference to the accompanying drawings and specific embodiments, which are intended to facilitate an understanding of the invention and are not intended to be limiting in any way.
The invention provides a method for identifying the insurance risk of the project performance guarantee based on characteristic engineering, which has the main flow as shown in figure 1 and comprises the following steps:
1. and carrying out preprocessing operation on the engineering service data, and constructing an initial training data set according to the preprocessed data.
The XGboost algorithm used in the invention cannot process character classification characteristics, so that the type characteristics need to be coded and converted, the characteristic structure diagram of the engineering performance assurance insurance field processed by the invention is shown in FIG. 3, in the embodiment, one-hot coding is used, the meaning is that N states are stored by using an N-bit register, each state has an independent register bit, and only one bit in the register is effective. For example, as shown in table 1, the "construction difficulty" feature includes three values, and thus can be expanded to three features. In the three-bit code after the original characteristic conversion, only the corresponding conversion bit is in the state 1, and the rest are 0, namely, the value of 'simple' can be converted into the code in which the values of 'construction difficulty _ simple', 'construction difficulty _ general' and 'construction difficulty _ complex' are respectively 1, 0 and 0.
TABLE 1 character quantity characteristic coding schematic table
Difficulty of construction Construction difficulty _ simple Construction difficulty _ general Construction difficulty _ Complex
Simple and easy 1 0 0
In general 0 1 0
Complexity of 0 0 1
In addition, the input engineering service information has partial missing values. In consideration of the actual meaning of data and the requirement of algorithm deployment, the median of the same feature dimension data can be used for filling a feature missing position, and the excessive influence on the data distribution and the actual meaning is avoided.
2. According to the initial training data set, a reference risk assessment model is obtained by utilizing XGboost model training, and the discrimination accuracy of the reference risk assessment model is recorded
Figure 695055DEST_PATH_IMAGE034
。
The XGboost (extreme Gradient Boosting) is an efficient implementation of a Gradient Boosting (GB) method, is a learning model for regression and classification problems, and has the characteristics of low probability of overfitting, high flexibility, high convergence speed, high accuracy and the like. The XGboost model is used, so that the risk assessment performance can be better. In the embodiment, the training data set obtained in the step 1 is used, an XGboost model with default parameters is used for direct training, a benchmark risk assessment model can be obtained, and the discrimination accuracy of the model is recorded at the moment
Figure 248527DEST_PATH_IMAGE002
For subsequent use. In observing the model results, the data results of the evaluation model have the following four possibilities:
a. true positive
Figure 542105DEST_PATH_IMAGE035
: the real type of the sample is positive, and the model prediction result is also positive;
b. true negative
Figure 989267DEST_PATH_IMAGE036
: the true category of the sample is negative, and the model prediction result is also negative;
c. false positive
Figure 851044DEST_PATH_IMAGE037
: the real type of the sample is negative, and the model prediction result is positive;
d. false negative
Figure 16446DEST_PATH_IMAGE038
: the true category of the sample is positive, and the model prediction result is negative.
The data related to the invention is classified data, and comprises two categories of 'application of insurance' and 'non-application of insurance'. The comparison standard of the model is mainly the model discrimination index of the "no-guarantee" data because the "no-guarantee" class data is less and the wrong discrimination of the classified data causes great loss to the company. If the "no guarantee" data used in the present invention is defined as positive class (Positive) "application" data is negative classNegative) Then the accuracy rate of the 'no guarantee' data can be calculatedPrecisionRecall rateRecall、F1-ScoreThe meaning is as follows:
a. rate of accuracyPrecision：
Figure 989081DEST_PATH_IMAGE039
The proportion of positive true categories in the data samples judged to be positive, namely the judgment accuracy of the model for the positive categories;
b. recall rateRecall：
Figure 290749DEST_PATH_IMAGE040
The proportion of the data samples with positive real categories judged to be positive;
c. F1-Score：
Figure 182482DEST_PATH_IMAGE041
F1-Scoreis a harmonic average of precision and recall.
In addition, the proportion of all samples which are judged to be correct is also required to be compared, namely the total accuracy:
Figure 710546DEST_PATH_IMAGE042
taken together, this embodiment uses
Figure 345927DEST_PATH_IMAGE003
Of the "non-insuring" typeRecallValue and model overall accuracy
Figure 767681DEST_PATH_IMAGE043
The sum of the values is used for achieving the purposes of considering the category data with greater threat to the service and considering the overall accuracy.
3. And performing feature screening by using an MR-MIC feature selection strategy and a benchmark risk evaluation model aiming at the initial training data set to obtain a screened training data set.
A. Generating mesh partitions
In practice, the parameters of mesh division need to be setBProduce a satisfactionmn<BVarious kinds of (A), (B), (Cm,n) A combination of positive integers of (a) is,Bif the parameter is too large, the number of mesh divisions is large, and calculation becomes complicated, and if the parameter is too small, the interval pattern of the division is too simple, and therefore, the parameter is generally set to be an empirical parameter
Figure 968331DEST_PATH_IMAGE044
。
B. Determining a maximum mutual information value
For each pair of characteristics in engineering service dataXAndYgo through each group (m,n) Will beXIs divided evenly intomShare and find the feature by using dynamic programmingXAndYfeatures with maximum mutual informationYIs then fixed to the featureYUsing dynamic programming to find the featuresXAndYfeatures with maximum mutual informationXIs divided, then, the feature is fixedXUsing dynamic programming to find the featuresXAndYfeatures with maximum mutual informationYAnd finally outputting each group of (m,n) Corresponding maximum mutual information valueI mn (X,Y)。
C. Determining maximum mutual information coefficient
Each pair is calculated according to the following formulaXAndYmaximum mutual information coefficient of
Figure 842746DEST_PATH_IMAGE007
：
Figure 16238DEST_PATH_IMAGE026
Method for calculating maximum mutual information coefficient between each feature and corresponding class label in engineering service data and each pair of featuresXAndYthe maximum mutual information coefficient calculation methods are consistent;
D. constructing an initial feature index set
Initially, all features are traversed
Figure 433444DEST_PATH_IMAGE045
Selecting the label of the category and the itemcThe largest mutual information coefficient is the largest, and an initial feature index set is constructed therefromS 1：
Figure 666979DEST_PATH_IMAGE009
Wherein
Figure 28691DEST_PATH_IMAGE010
For the first in engineering business datakThe characteristics of the device are as follows,cis a category label;
Figure 146819DEST_PATH_IMAGE011
for the features calculated according to step S32 and step S33
Figure 543166DEST_PATH_IMAGE010
And its corresponding category labelcMaximum mutual information coefficient therebetween;
E. constructing all feature index sets
After the initial feature index set is obtained, one and the category label are selected for each feature additioncThe index of the feature with the highest correlation and the lowest correlation with the selected features, and the rest feature index set is generated by the following formula
Figure 213181DEST_PATH_IMAGE012
：
Figure 671976DEST_PATH_IMAGE046
WhereinTRepresenting the total number of features in the engineering business data;
Figure 718429DEST_PATH_IMAGE047
indexing sets for featuresS t Is indexed byiIs characterized in that it is a mixture of two or more of the above-mentioned components,
Figure 969282DEST_PATH_IMAGE015
indexing sets for unselected features
Figure 419986DEST_PATH_IMAGE048
The middle index isjThe characteristics of (1).
F. Performing model judgment and result recording
After all feature index sets are generated, each feature index set needs to be generatedS t Inputting the corresponding data set into the XGboost model, and recording the discrimination accuracy
Figure 756289DEST_PATH_IMAGE003
And selecting the feature index set with the highest discrimination accuracy
Figure 75275DEST_PATH_IMAGE049
Record the highest discrimination accuracy
Figure 56001DEST_PATH_IMAGE004
。
G. Feature index set selection
Will be provided with
Figure 67819DEST_PATH_IMAGE004
The discrimination accuracy of the reference risk assessment model in the step 2
Figure 891419DEST_PATH_IMAGE002
Make a comparison if
Figure 889462DEST_PATH_IMAGE050
Then determine
Figure 114906DEST_PATH_IMAGE017
In the embodiment, the screening of the optimal feature index set can be completed through the standard. In addition, when
Figure 32047DEST_PATH_IMAGE051
If the results show different losses after screening, the process needs to be traversed from large to smalltIn an embodiment, the setting finds a satisfaction
Figure 218309DEST_PATH_IMAGE052
All are the same asNumber of features that are to be screened out
Figure 879097DEST_PATH_IMAGE053
Characteristic index set ofS t I.e., accuracy does not decrease by more than 5% and more than 20% of the features are screened out and determined to be the final selected feature index set, the selection criteria being used to achieve the goal of deleting as many features as possible while preserving data performance.
H. Obtaining a filtered data set
And screening the engineering service data by using the finally selected feature index set so as to obtain a screened training data set.
4. And aiming at the screened training data set, training by using an XGboost model to obtain a final risk assessment model.
In this embodiment, after the final feature index set and the filtered data set are determined, the model is trained again by using the filtered data, and the comparison between the "non-insurable" model indexes before and after feature filtering and the accuracy is shown in table 2:
TABLE 2 comparison of "No insurances" class model indices before and after feature screening with accuracy
Precision Recall F1-Score Accuracy
Before screening 0.67 0.55 0.61 0.86
After screening 0.71 0.56 0.63 0.87
The observation shows that after the characteristic screening, the model index of the 'no-guarantee' class is obviously improved, and the overall accuracy rate is increased, which shows that the MR-MIC characteristic screening method has better effect.
5. And (3) after the data of the project to be evaluated is subjected to preprocessing operation in the step (1) to obtain project service data, using the characteristic screening in the step (3), and then inputting the data subjected to preprocessing and characteristic screening into the final risk evaluation model obtained in the step (4) to obtain a risk identification result of the project to be evaluated.
As shown in FIG. 2, the invention also provides a feature engineering-based engineering performance guarantee insurance risk identification platform, which comprises a data input module, a data processing module, a feature calculation and screening module, a model training module and a risk assessment module
The data input module is used for receiving engineering project information data needing risk identification, and the data input module comprises engineering project information data input for model training or engineering project information data to be evaluated;
the data processing module is used for executing preprocessing operation on the engineering project information data to obtain engineering service data, and generating an initial training data set or preprocessing the engineering project information data to be evaluated;
the characteristic calculation and screening module is used for carrying out characteristic screening on the initial training data set processed by the data processing module by utilizing characteristic engineering, the characteristic engineering is a maximum correlation minimum redundancy combined maximum mutual information coefficient characteristic selection strategy, which is recorded as MR-MIC, and a reference risk discrimination model obtained by combining the model training module and the discrimination accuracy rate thereof
Figure 959049DEST_PATH_IMAGE001
Firstly, calculating the maximum mutual information coefficient of each pair of characteristics and each characteristic and the corresponding class label in engineering service data, then constructing a characteristic index set, and recording the judgment accuracy rate of each characteristic index set
Figure 185106DEST_PATH_IMAGE003
Selecting the feature index set with the highest accuracy, and recording the highest discrimination accuracy
Figure 717719DEST_PATH_IMAGE004
Accuracy of discrimination from reference risk assessment model
Figure 182198DEST_PATH_IMAGE034
For comparison, if
Figure 257601DEST_PATH_IMAGE054
Then determining the selected feature index set as the finally selected feature index set, if so
Figure 516544DEST_PATH_IMAGE006
Then, the feature index set is sorted and traversed from large to small according to the feature number in the feature index set, and a feature index set is found, and the judgment accuracy rate is high
Figure 536453DEST_PATH_IMAGE003
Greater than a threshold of accuracy
Figure 945569DEST_PATH_IMAGE052
The accuracy threshold value is used for judging the accuracy according to
Figure 734533DEST_PATH_IMAGE055
And the required precision is selected, and the requirement that the number of the screened features is larger than the threshold value of the number of the features is met
Figure 305323DEST_PATH_IMAGE023
(ii) a Performing feature screening based on the feature index set to obtain a screened training data set;
the model training module is used for training the initial training data set processed by the data processing module by using the XGboost model to obtain a reference risk discrimination model and recording the discrimination accuracy of the model
Figure 812528DEST_PATH_IMAGE056
Or training the screened training data set generated by the feature calculation and screening module by using an XGboost model to obtain a final risk discrimination model;
and the risk evaluation module is used for giving a risk judgment result of the information data of the engineering project to be evaluated, which is input by the data input module, according to the final risk evaluation model.
The present invention is not limited to the above-described embodiments, and those skilled in the art can implement the present invention in other various embodiments based on the disclosure of the present invention. Therefore, the design of the invention is within the scope of protection, with simple changes or modifications, based on the design structure and thought of the invention.

Claims (5)

1. A method for identifying the risk of insurance for ensuring the performance of an engineering based on characteristic engineering is characterized by comprising the following steps:
s1: carrying out preprocessing operation on the engineering project information data to obtain engineering service data, and constructing an initial training data set according to the engineering service data;
s2: training by using an XGboost model according to an initial training data setObtaining a reference risk evaluation model and recording the discrimination accuracy of the reference risk evaluation model
Figure DEST_PATH_IMAGE002
；
S3: and (3) performing feature screening by using feature engineering aiming at an initial training data set, wherein the feature engineering is a maximum correlation minimum redundancy combined maximum mutual information coefficient feature selection strategy, is recorded as MR-MIC, and is combined with a reference risk assessment model and the judgment accuracy rate thereof
Figure DEST_PATH_IMAGE003
Obtaining a screened training data set; the method specifically comprises the following steps:
s31: setting a mesh partition size parameterBProduce a satisfactionm*n<BVarious kinds of (A), (B), (Cm,n) A combination of positive integers of (a) is,mandnvalues for grid horizontal and vertical division;
s32: for each pair of characteristics in engineering service dataXAndYgo through each group (m,n) Will beXIs divided evenly intomShare and find the feature by using dynamic programmingXAndYfeatures with maximum mutual informationYIs then fixed to the featureYUsing dynamic programming to find the featuresXAndYfeatures with maximum mutual informationXIs divided, then, the feature is fixedXUsing dynamic programming to find the featuresXAndYfeatures with maximum mutual informationYAnd finally outputting each group of (m,n) Corresponding maximum mutual information valueI mn (X,Y)；
S33: each pair is calculated according to the following formulaXAndYmaximum mutual information coefficient of
Figure DEST_PATH_IMAGE005
：
Figure DEST_PATH_IMAGE007
Method for calculating maximum mutual information coefficient between each feature and corresponding class label in engineering service data and each pair of featuresXAndYthe maximum mutual information coefficient calculation methods are consistent;
s34: constructing feature index setsS 1：
Figure DEST_PATH_IMAGE009
Wherein
Figure DEST_PATH_IMAGE011
For the first in engineering business datakThe characteristics of the device are as follows,cis a category label;
Figure DEST_PATH_IMAGE013
for the features calculated according to step S32 and step S33
Figure 609685DEST_PATH_IMAGE011
And its corresponding category labelcMaximum mutual information coefficient therebetween;
s35: generating the remaining feature index set by the following formula
Figure DEST_PATH_IMAGE015
：
Figure DEST_PATH_IMAGE017
WhereinTRepresenting the total number of features in the engineering business data;
Figure DEST_PATH_IMAGE019
indexing sets for featuresS t Is indexed byiIs characterized in that it is a mixture of two or more of the above-mentioned components,
Figure DEST_PATH_IMAGE021
indexing sets for unselected features
Figure DEST_PATH_IMAGE023
The middle index isjThe features of (1);
s36: indexing each feature into a setS t Inputting the corresponding data set into the XGboost model, and recording the discrimination accuracy
Figure DEST_PATH_IMAGE025
And selecting the feature index set with the highest accuracy
Figure DEST_PATH_IMAGE027
Simultaneously recording the highest discrimination accuracy
Figure DEST_PATH_IMAGE029
；
S37: will be provided with
Figure 594434DEST_PATH_IMAGE029
Determination accuracy of the reference risk assessment model in step S2
Figure DEST_PATH_IMAGE030
Make a comparison if
Figure DEST_PATH_IMAGE032
Then determine
Figure 132863DEST_PATH_IMAGE027
For the finally selected feature index set, if
Figure DEST_PATH_IMAGE034
Then go from big to smalltFind onetThe accuracy of the discrimination
Figure 737150DEST_PATH_IMAGE025
Greater than a threshold of accuracy, i.e.
Figure DEST_PATH_IMAGE036
And the number of features to be screened out is greater than the threshold number of features, i.e. the number of features to be screened out
Figure DEST_PATH_IMAGE038
And determineS t As a final selected feature index set, wherein,aandbis a parameter set according to requirements;
s38: performing feature screening based on the finally selected feature index set to obtain a screened training data set;
s4: aiming at the screened training data set, using an XGboost model to train to obtain a final risk assessment model;
s5: after the engineering business data is obtained by the preprocessing operation in the step S1 on the information data of the engineering project to be evaluated, the MR-MIC feature screening in the step S3 is used, and then the engineering business data after preprocessing and feature screening is input to the final risk evaluation model obtained in the step S4, so as to obtain a risk evaluation result of the project to be evaluated.
2. The method as claimed in claim 1, wherein the preprocessing operation in step S1 includes:
and carrying out one-hot coding processing on the class characteristics described in the form of characters in the engineering service data to obtain discrete numerical characteristics, and meanwhile, filling missing values in the characteristics described in the form of numerical values in the engineering service data by using a median filling method to finish data preprocessing.
3. A project performance guarantee insurance risk identification platform based on feature engineering is characterized by comprising a data input module, a data processing module, a feature calculation and screening module, a model training module and a risk assessment module:
the data input module is used for receiving engineering project information data needing risk identification, and the data input module comprises engineering project information data input for model training or engineering project information data to be evaluated;
the data processing module is used for executing preprocessing operation on the engineering project information data to obtain engineering service data, and generating an initial training data set or preprocessing the engineering project information data to be evaluated;
the characteristic calculation and screening module is used for carrying out characteristic screening on the initial training data set processed by the data processing module by utilizing characteristic engineering, the characteristic engineering is a maximum correlation minimum redundancy combined maximum mutual information coefficient characteristic selection strategy, which is recorded as MR-MIC, and a reference risk discrimination model obtained by combining the model training module and the discrimination accuracy rate thereof
Figure DEST_PATH_IMAGE039
And (3) carrying out feature screening to obtain a screened training data set, which specifically comprises the following steps:
the characteristic calculating and screening module comprises a maximum mutual information coefficient calculating module, a characteristic index set generating module and a characteristic screening module;
the maximum mutual information coefficient calculation module is used for calculating each pair of characteristics in the engineering service data obtained by the data processing moduleXAndYor the maximum mutual information coefficient between each feature and its corresponding class label; the method comprises the following specific steps:
(1) setting a mesh partition size parameterBProduce a satisfactionm*n<BVarious kinds of (A), (B), (Cm,n) A combination of positive integers of (a) is,mandnvalues for grid horizontal and vertical division;
(2) for each pair of characteristics in engineering service dataXAndYgo through each group (m,n) Will beXIs divided evenly intomShare and find the feature by using dynamic programmingXAndYfeatures with maximum mutual informationYIs then fixed to the featureYUsing dynamic programming to find the featuresXAndYthe most mutual information between themSign forXIs divided, then, the feature is fixedXUsing dynamic programming to find the featuresXAndYfeatures with maximum mutual informationYAnd finally outputting each group of (m,n) Corresponding maximum mutual information valueI mn (X,Y)；
(3) Each pair is calculated according to the following formulaXAndYmaximum mutual information coefficient of
Figure 550386DEST_PATH_IMAGE005
：
Figure 454888DEST_PATH_IMAGE007
Method for calculating maximum mutual information coefficient between each feature and corresponding class label in engineering service data and each pair of featuresXAndYthe maximum mutual information coefficient calculation methods are consistent;
the feature index set generation module is configured to perform feature screening on the data preprocessed by the data processing module by using an MR-MIC feature selection policy according to the maximum mutual information coefficients between each pair of features calculated by the maximum mutual information coefficient module and between each feature and the corresponding category label thereof, and generate all feature index sets, specifically as follows:
(1) constructing feature index setsS 1：
Figure 757693DEST_PATH_IMAGE009
Wherein
Figure 114856DEST_PATH_IMAGE011
For the first in engineering business datakThe characteristics of the device are as follows,cis a category label;
Figure 731782DEST_PATH_IMAGE013
is based onFeatures obtained by maximum mutual information coefficient calculation module
Figure 615425DEST_PATH_IMAGE011
And its corresponding category labelcMaximum mutual information coefficient therebetween;
(2) generating the remaining feature index set by the following formula
Figure 230077DEST_PATH_IMAGE015
：
Figure 402432DEST_PATH_IMAGE017
WhereinTRepresenting the total number of features in the engineering business data;
Figure 619787DEST_PATH_IMAGE019
indexing sets for featuresS t Is indexed byiIs characterized in that it is a mixture of two or more of the above-mentioned components,
Figure 495952DEST_PATH_IMAGE021
indexing sets for unselected features
Figure 140560DEST_PATH_IMAGE023
The middle index isjThe features of (1);
the feature screening module is used for selecting the feature index set with the highest accuracy value from all the feature index sets obtained by the feature index set generation module
Figure 800211DEST_PATH_IMAGE027
Simultaneously recording the highest discrimination accuracy
Figure 431044DEST_PATH_IMAGE029
Accuracy of discrimination from reference risk assessment model
Figure DEST_PATH_IMAGE040
For comparison, if
Figure DEST_PATH_IMAGE041
Then determine
Figure 695803DEST_PATH_IMAGE027
For the finally selected feature index set, if
Figure DEST_PATH_IMAGE042
Then go from big to smalltFind onetThe accuracy of the discrimination
Figure 917837DEST_PATH_IMAGE025
Greater than a threshold of accuracy, i.e.
Figure DEST_PATH_IMAGE043
And the number of features to be screened out is greater than the threshold number of features, i.e. the number of features to be screened out
Figure 861522DEST_PATH_IMAGE038
And determineS t As a final selected feature index set, wherein,aandbthe method comprises the steps of performing feature screening on parameters set according to requirements and based on a finally selected feature index set to obtain a screened training data set;
the model training module is used for training the initial training data set processed by the data processing module by using the XGboost model to obtain a reference risk discrimination model and recording the discrimination accuracy of the reference risk discrimination model
Figure DEST_PATH_IMAGE044
(ii) a Or training the screened training data set generated by the feature calculation and screening module by using an XGboost model to obtain a final risk discrimination model;
and the risk evaluation module is used for giving a risk judgment result of the information data of the engineering project to be evaluated, which is input by the data input module, according to the final risk evaluation model.
4. The platform of claim 3, wherein the data input module comprises a database for receiving data input from outside in a unified manner.
5. The feature engineering-based project performance guarantee insurance risk identification platform according to claim 3, wherein the data processing module comprises a word feature processing module and a numerical feature processing module;
the character feature processing module is used for carrying out one-hot coding processing on the class features described in the form of characters in the engineering service data to obtain discrete numerical features;
and the numerical value characteristic processing module is used for filling missing values by using a median filling method aiming at the characteristics described in a numerical value form in the engineering service data.
CN202010739603.9A 2020-07-28 2020-07-28 Method and platform for identifying insurance risk of engineering performance guarantee based on characteristic engineering Active CN111738870B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010739603.9A CN111738870B (en) 2020-07-28 2020-07-28 Method and platform for identifying insurance risk of engineering performance guarantee based on characteristic engineering

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010739603.9A CN111738870B (en) 2020-07-28 2020-07-28 Method and platform for identifying insurance risk of engineering performance guarantee based on characteristic engineering

Publications (2)

Publication Number Publication Date
CN111738870A CN111738870A (en) 2020-10-02
CN111738870B true CN111738870B (en) 2020-12-25

Family

ID=72656242

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010739603.9A Active CN111738870B (en) 2020-07-28 2020-07-28 Method and platform for identifying insurance risk of engineering performance guarantee based on characteristic engineering

Country Status (1)

Country Link
CN (1) CN111738870B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112529716B (en) * 2020-12-07 2023-10-24 平安科技（深圳）有限公司 Method, device and computer readable storage medium for predicting credit
CN113159568A (en) * 2021-04-19 2021-07-23 福建万川信息科技股份有限公司 System and method for estimating insurance risk

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170193335A1 (en) * 2015-11-13 2017-07-06 Wise Athena Inc. Method for data encoding and accurate predictions through convolutional networks for actual enterprise challenges
CN108509388A (en) * 2018-01-30 2018-09-07 天津大学 Feature selection approach based on maximal correlation minimal redundancy and sequence
CN111401914B (en) * 2020-04-02 2022-07-22 支付宝(杭州)信息技术有限公司 Risk assessment model training and risk assessment method and device

Also Published As

Publication number Publication date
CN111738870A (en) 2020-10-02

Similar Documents

Publication Publication Date Title
CN109829631B (en) Enterprise risk early warning analysis method and system based on memory network
McKee Rough sets bankruptcy prediction models versus auditor signalling rates
CN108459955B (en) Software defect prediction method based on deep self-coding network
CN111738870B (en) Method and platform for identifying insurance risk of engineering performance guarantee based on characteristic engineering
Silva et al. Cross country relations in European tourist arrivals
Ko et al. Prediction of corporate financial distress: An application of the composite rule induction system
CN112700319A (en) Enterprise credit line determination method and device based on government affair data
CN112559900B (en) Product recommendation method and device, computer equipment and storage medium
CN113139687B (en) Method and device for predicting credit card user default
CN111104975B (en) Credit evaluation method based on breadth learning
CN114048436A (en) Construction method and construction device for forecasting enterprise financial data model
CN112613977A (en) Personal credit loan admission credit granting method and system based on government affair data
CN111754317A (en) Financial investment data evaluation method and system
CN114926299A (en) Prediction method for predicting vehicle accident risk based on big data analysis
CN116340726A (en) Energy economy big data cleaning method, system, equipment and storage medium
Leqi et al. Supervised learning with general risk functionals
CN1653486B (en) Pattern feature selection method, classification method, judgment method, program, and device
CN117689209A (en) Bridge security risk assessment method, equipment and medium based on entropy weight fuzzy synthesis
CN111626886A (en) Multi-party cooperation-based engineering performance guarantee insurance risk identification method and platform
Hamidieh Estimating the tail shape parameter from option prices
CN112766765A (en) Professional learning ability evaluation method and system based on interval middle intelligence theory
CN114170000A (en) Credit card user risk category identification method, device, computer equipment and medium
CN113191771A (en) Buyer account period risk prediction method
CN114386647A (en) Method and system for predicting energy consumption of oil and gas field industry
CN111915428A (en) Scoring card model optimization method based on decision tree feature fusion

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant