CN109754281A - A kind of supplier's attrition prediction method - Google Patents
A kind of supplier's attrition prediction method Download PDFInfo
- Publication number
- CN109754281A CN109754281A CN201811397492.7A CN201811397492A CN109754281A CN 109754281 A CN109754281 A CN 109754281A CN 201811397492 A CN201811397492 A CN 201811397492A CN 109754281 A CN109754281 A CN 109754281A
- Authority
- CN
- China
- Prior art keywords
- class
- cluster
- supplier
- data
- neural network
- 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.)
- Granted
Links
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention relates to a kind of supplier's attrition prediction methods, first according to the demand of practical problem, are integrated to the data of platform itself, determine the feature for being lost supplier；Secondly unbalanced dataset is sampled using MBCDK-means lack sampling method, unbalanced data is converted into equilibrium data collection；Then Genetic Artificial Neural Network method is utilized, equilibrium data collection is predicted；Finally, output prediction result.The present invention can carry out Accurate Prediction to the loss of supplier.
Description
Technical field
The present invention relates to supplier's attrition prediction technical fields, more particularly to a kind of supplier's attrition prediction method.
Background technique
Customer churn prediction is a major issue in customer relation management.In increasingly competitive business environment now
In, client can easily switch between rival.It is some studies have shown that obtain the cost of new client usually more existing than retaining
There are expensive 5 to 6 times of the cost of client.Meanwhile long-term customers are lower to the susceptibility of competitive marketing activity, profit is higher.In addition,
The loss of client not only results in revenue losses, also results in brand loyalty decline and influences the morale of company.Therefore, public
Emphasis is transferred to reservation existing customer group from new client is developed by department.Accurate customer churn prediction, which will facilitate company, to be closed
Suitable client navigates in reserved-range, therefore is acknowledged as the marketing matter of priority.
Customer churn means that client's subsidiary company cancels service, and the existing customer for retaining enterprise plays an important role, to increase
Add the overall income of company, and the company status being retained in market with keen competition.The reason of leading to customer churn, has very much,
So determining these the reason is that considerably complicated, because they depend on personal view and the company that is utilizing of client of client
Service, but determine customer churn be necessary again.
Summary of the invention
Technical problem to be solved by the invention is to provide a kind of supplier's attrition prediction methods, can be to the stream of supplier
It loses and carries out Accurate Prediction.
The technical solution adopted by the present invention to solve the technical problems is: providing a kind of supplier's attrition prediction method, wraps
Include following steps:
(1) supplier of acquisition reflection platform is lost the related data of feature；
(2) collected unbalanced dataset is divided into uneven training dataset and uneven test data set；
(3) balance training data set is converted for uneven training dataset using MBCDK-means lack sampling method；
(4) prediction model is established using Genetic Artificial Neural Network method；
(5) uneven test data set collection is predicted using the prediction model, exports prediction result.
It includes: that qualification certificates, company's type, registered capital, registration information are complete that supplier, which is lost feature, in the step (1)
It is whole degree, concern bidding documents number, nearest attitude, company's qualification, service quality, product quality, delivery rate, credibility, in
Mark number, bid number, agreed-upon price number, login times, reasonable price degree and contract agreement fulfillment rate.
Further include that pretreated step is carried out to the data of acquisition between the step (1) and step (2), specifically include:
Data integration is carried out to data；Data are cleaned, including removing noise and deleting inconsistent data；Data are become
It changes, including construction new feature and data normalization.
The step (3) specifically includes following sub-step:
(31) uneven training dataset is divided into M most class samples and N number of minority class sample；
(32) initialization cluster number K；
(33) K class is polymerized to using K-means algorithm to M most class samples, a kind of composition is polymerized to N number of minority class sample
Minority class subset；
(34) cluster centre of classes most for i-th, calculates the distance of its cluster centre for arriving minority classWherein, XiIndicate the cluster center of i-th of cluster, XNIndicate the cluster center of minority class；
(35) average distance of the most class cluster centers of calculating to minority class cluster center
(36) sample is selected to constitute most class subsets from each cluster of most classes, wherein sample size ismiIndicate the sample number in the cluster of i-th of most class cluster；
(37) most class subsets and minority class subset are constituted into balance training data set.
The step (4) specifically includes following sub-step:
(41) n individual of the first generation is randomly generated；
(42) n neural network is initialized；
(43) training neural network；
(44) whether the neural network after training of judgement reaches setting target, goes to step (45) if not reaching, otherwise turns
Step (46)；
(45) genetic operation is carried out, duplication, intersection and variation including chromosome obtain n individual of new generation, and return
Step (42)；
(46) optimal neural network is selected；
(47) according to optimal neural network Genetic Neural Network Predictive Model.
Beneficial effect
Due to the adoption of the above technical solution, compared with prior art, the present invention having the following advantages that and actively imitating
Fruit: the present invention is selected with most classes to the distance at minority class cluster center when sampling according to the sample distribution quantity in cluster
Number of samples is taken, retains the distributed intelligence of initial data cluster and improves boundary sample sample rate simultaneously, help to improve final classification
Performance；The present invention uses MBCDK-means lack sampling method, reduces time and the space of existing K-means lack sampling algorithm
Complexity；Present invention introduces weights and biasing that genetic algorithm carrys out optimized artificial neural network, construct genetic neural network mould
Type has better estimated performance compared to artificial neural network.
Detailed description of the invention
Fig. 1 is flow chart of the invention；
Fig. 2 is to convert balance instruction for uneven training dataset using MBCDK-means lack sampling method in the present invention
Practice the flow chart of data set；
Fig. 3 is the flow chart for establishing prediction model in the present invention using Genetic Artificial Neural Network method.
Specific embodiment
Present invention will be further explained below with reference to specific examples.It should be understood that these embodiments are merely to illustrate the present invention
Rather than it limits the scope of the invention.In addition, it should also be understood that, after reading the content taught by the present invention, those skilled in the art
Member can make various changes or modifications the present invention, and such equivalent forms equally fall within the application the appended claims and limited
Range.
Embodiments of the present invention are related to a kind of supplier's attrition prediction method, first according to the demand of practical problem, knot
The data of platform itself are closed, determine the feature for being lost supplier；Secondly using MBCDK-means lack sampling method to imbalance
Data set is sampled, and unbalanced data is converted to equilibrium data collection；Then Genetic Artificial Neural Network method is utilized, to flat
Weighing apparatus data set is predicted；Finally, output prediction result.As shown in Figure 1, the specific steps of which are as follows:
Step A, according to the demand of practical problem, the data of platform itself are integrated to, determine the loss feature of supplier；This
Embodiment is for building supplier, wherein the loss feature of determining supplier includes: qualification certificates, company's type, note
Volume fund, registration information integrity degree, pay close attention to bidding documents number and nearest attitude (such as nearest two months, four months,
Half a year), company's qualification, service quality, product quality, delivery rate, credibility, acceptance of the bid number, bid number, agreed-upon price number,
Login times, reasonable price degree, contract agreement fulfillment rate.
Step B, reflect that supplier is lost the related data of feature on acquisition platform；
Step C, data prediction is carried out to the data of acquisition, specifically included；
C1, data integration is carried out to data；
C2, data are cleaned, including removing noise and deleting inconsistent data；
C3, data are converted, including construction new feature and data normalization.
Step D, unbalanced dataset is divided into uneven training dataset and uneven test data set；
Step E, balance training data are converted for uneven training dataset using MBCDK-means lack sampling method
Collection, as shown in Fig. 2, specifically including:
E1, training set is divided into M most class samples and N number of minority class sample；
E2, initialization cluster number K；
E3, K class is polymerized to using K-means algorithm to most class samples；
E4, a kind of composition minority class subset is polymerized to minority class sample；
The cluster centre of E5, class most for i-th, calculate the distance of its cluster centre for arriving minority classWherein XiIndicate the cluster center of i-th of cluster, XNIndicate the cluster center of minority class；
E6, calculate most class cluster centers to minority class cluster center average distance
E7, for i-th of most class, the sample number selected in suchmiIndicate i-th of most class
Sample number in the cluster of cluster；
E8, the most class subsets of sampling composition are carried out from each most classes according to sample number；
E9, most class subsets and minority class subset are constituted into balance training collection.
Step F, using Genetic Artificial Neural Network method, prediction model is established；
F1, n individual of the first generation is randomly generated；
N F2, initialization neural network；
F3, training neural network；
F4, judge whether to reach setting target, go to step F5 if not reaching, otherwise go to step F8；
F5, chromosome replication；
F6, chiasma；
F7, chromosomal variation, go to step F2；
F8, selection optimal neural network；
F9, Genetic Neural Network Predictive Model is obtained.
Model parameter in present embodiment is provided that
| The number of iterations | 1000 |
| Learning rate | 0.05 |
| Target error | 0.0001 |
| Population Size N | 40 |
| Evolutionary generation T | 100 |
| Crossover probability Pc | 0.8 |
| Mutation probability Pm | 0.02 |
Step G, test set is predicted, exports prediction result.
The present invention is when sampling, according to the distance of sample distribution quantity and most classes to minority class cluster center in cluster
It chooses number of samples, retains the distributed intelligence of initial data cluster and improve boundary sample sample rate simultaneously, help to improve final
Classification performance；The present invention use MBCDK-means lack sampling method, reduce existing K-means lack sampling algorithm time and
Space complexity；Present invention introduces weights and biasing that genetic algorithm carrys out optimized artificial neural network, construct genetic neural network
Network model has better estimated performance compared to artificial neural network.
Claims (5)
1. a kind of supplier's attrition prediction method, which comprises the following steps:
(1) supplier of acquisition reflection platform is lost the related data of feature；
(2) collected unbalanced dataset is divided into uneven training dataset and uneven test data set；
(3) balance training data set is converted for uneven training dataset using MBCDK-means lack sampling method；
(4) prediction model is established using Genetic Artificial Neural Network method；
(5) uneven test data set collection is predicted using the prediction model, exports prediction result.
2. supplier's attrition prediction method according to claim 1, which is characterized in that supply commodity-circulate in the step (1)
Losing feature includes: qualification certificates, company's type, registered capital, registration information integrity degree, concern bidding documents number, nearest service state
Degree, acceptance of the bid number, bid number, agreed-upon price number, logs in company's qualification, service quality, product quality, delivery rate, credibility
Number, reasonable price degree and contract agreement fulfillment rate.
3. supplier's attrition prediction method according to claim 1, which is characterized in that the step (1) and step (2) it
Between further include that pretreated step is carried out to the data of acquisition, specifically include: data integration carried out to data；Data are carried out clear
It washes, including removing noise and deleting inconsistent data；Data are converted, including construction new feature and data normalizing
Change.
4. supplier's attrition prediction method according to claim 1, which is characterized in that the step (3) specifically include with
Lower sub-step:
(31) uneven training dataset is divided into M most class samples and N number of minority class sample；
(32) initialization cluster number K；
(33) K class is polymerized to using K-means algorithm to M most class samples, it is a small number of to be polymerized to a kind of composition to N number of minority class sample
Class subset；
(34) cluster centre of classes most for i-th, calculates the distance of its cluster centre for arriving minority classWherein, XiIndicate the cluster center of i-th of cluster, XNIndicate the cluster center of minority class；
(35) average distance of the most class cluster centers of calculating to minority class cluster center
(36) sample is selected to constitute most class subsets from each cluster of most classes, wherein sample size ismiIndicate the sample number in the cluster of i-th of most class cluster；
(37) most class subsets and minority class subset are constituted into balance training data set.
5. supplier's attrition prediction method according to claim 1, which is characterized in that the step (4) specifically include with
Lower sub-step:
(41) n individual of the first generation is randomly generated；
(42) n neural network is initialized；
(43) training neural network；
(44) whether the neural network after training of judgement reaches setting target, goes to step (45) if not reaching, otherwise goes to step
(46)；
(45) genetic operation is carried out, duplication, intersection and variation including chromosome obtain n individual of new generation, and return step
(42)；
(46) optimal neural network is selected；
(47) according to optimal neural network Genetic Neural Network Predictive Model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811397492.7A CN109754281B (en) | 2018-11-22 | 2018-11-22 | Supplier loss prediction method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811397492.7A CN109754281B (en) | 2018-11-22 | 2018-11-22 | Supplier loss prediction method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109754281A true CN109754281A (en) | 2019-05-14 |
| CN109754281B CN109754281B (en) | 2021-11-19 |
Family
ID=66402533
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811397492.7A Active CN109754281B (en) | 2018-11-22 | 2018-11-22 | Supplier loss prediction method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109754281B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112862546A (en) * | 2021-04-25 | 2021-05-28 | 平安科技（深圳）有限公司 | User loss prediction method and device, computer equipment and storage medium |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107944460A (en) * | 2016-10-12 | 2018-04-20 | 甘肃农业大学 | One kind is applied to class imbalance sorting technique in bioinformatics |
-
2018
- 2018-11-22 CN CN201811397492.7A patent/CN109754281B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107944460A (en) * | 2016-10-12 | 2018-04-20 | 甘肃农业大学 | One kind is applied to class imbalance sorting technique in bioinformatics |
Non-Patent Citations (1)
| Title |
|---|
| 刘晨晨: "基于数据挖掘的通信客户流失预警模型研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112862546A (en) * | 2021-04-25 | 2021-05-28 | 平安科技（深圳）有限公司 | User loss prediction method and device, computer equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109754281B (en) | 2021-11-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109034915B (en) | Artificial intelligent electronic commerce system capable of using digital assets or points as transaction media | |
| Corazza et al. | Particle Swarm Optimization with non-smooth penalty reformulation, for a complex portfolio selection problem | |
| Lin et al. | Tourism demand forecasting: Econometric model based on multivariate adaptive regression splines, artificial neural network and support vector regression | |
| CN112926651A (en) | Enterprise credit assessment method and system | |
| CN107609771A (en) | A kind of supplier's value assessment method | |
| CN110826886A (en) | Electric power customer portrait construction method based on clustering algorithm and principal component analysis | |
| CN110298574A (en) | A kind of electricity consumption subscriber payment risk rating method based on convolutional neural networks | |
| CN108230029A (en) | Client trading behavior analysis method | |
| CN110866782A (en) | Customer classification method and system and electronic equipment | |
| CN107481135A (en) | A kind of personal credit evaluation method and system based on BP neural network | |
| CN114943565A (en) | Electric power spot price prediction method and device based on intelligent algorithm | |
| Li et al. | Predicting business risks of commercial banks based on BP-GA optimized model | |
| CN115423538A (en) | Method and device for predicting new product sales data, storage medium and electronic equipment | |
| CN109754281A (en) | A kind of supplier's attrition prediction method | |
| CN112163781A (en) | Park electricity utilization group life cycle evaluation method based on multi-dimensional index clustering | |
| CN109190820B (en) | Electric power market electricity selling quantity depth prediction method considering user loss rate | |
| CN116611911A (en) | Credit risk prediction method and device based on support vector machine | |
| CN110807543A (en) | Investment portfolio optimization method and device based on group decision intelligent search | |
| Cao | Research on the impact of artificial intelligence-based e-commerce personalization on traditional accounting methods | |
| Sharma et al. | Prediction of Real-Time Estate Pricing using Train-Test Splitting Techniques | |
| Gonçalves et al. | Credit risk analysis applying logistic regression, neural networks and genetic algorithms models | |
| CN111667307A (en) | Method and device for predicting financial product sales volume | |
| Wei et al. | A new dynamic credit scoring model based on clustering ensemble | |
| Ohsato et al. | Construction of an Input-Output Table Considering Business-to-Consumer Transactions by using Private Data | |
| Wang | Credit Strategy Design of Small and Medium-Sized Enterprises |
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 |