CN109190796A - A kind of telecom client attrition prediction method, system and electronic equipment - Google Patents
A kind of telecom client attrition prediction method, system and electronic equipment Download PDFInfo
- Publication number
- CN109190796A CN109190796A CN201810871287.3A CN201810871287A CN109190796A CN 109190796 A CN109190796 A CN 109190796A CN 201810871287 A CN201810871287 A CN 201810871287A CN 109190796 A CN109190796 A CN 109190796A
- Authority
- CN
- China
- Prior art keywords
- client
- network
- value
- sample
- node
- 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
- 238000000034 method Methods 0.000 title claims abstract description 60
- 238000007637 random forest analysis Methods 0.000 claims abstract description 86
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 65
- 230000006872 improvement Effects 0.000 claims abstract description 26
- 238000000638 solvent extraction Methods 0.000 claims description 30
- 238000004891 communication Methods 0.000 claims description 17
- 238000013138 pruning Methods 0.000 claims description 11
- 238000012360 testing method Methods 0.000 claims description 6
- 238000005192 partition Methods 0.000 claims description 5
- 238000004590 computer program Methods 0.000 claims description 4
- 238000001514 detection method Methods 0.000 claims description 4
- 102100034286 Ankyrin repeat domain-containing protein 27 Human genes 0.000 claims description 3
- 101000780114 Homo sapiens Ankyrin repeat domain-containing protein 27 Proteins 0.000 claims description 3
- 230000008569 process Effects 0.000 description 8
- 238000004364 calculation method Methods 0.000 description 5
- 238000012549 training Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 230000001052 transient effect Effects 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000003066 decision tree Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000007257 malfunction Effects 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 239000012535 impurity Substances 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000004579 scanning voltage microscopy Methods 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 238000005303 weighing Methods 0.000 description 1
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
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G06Q50/40—
Abstract
The present invention provides a kind of telecom client attrition prediction method, system and electronic equipment, the described method includes: calculating in telecommunication network at netter family and the corresponding sample division proportion of off-network client, and the improvement Geordie radix of random forests algorithm is determined at netter family and the corresponding sample division proportion of off-network client based on described；Based on the improvement Geordie radix, using random forests algorithm, predict that telecom client is lost；Wherein, the sample division proportion indicates, the correspondence classification of selection accounts for the category in the ratio of net state client's total amount in the sample size of the client of net state.The present invention can effectively solve the problem that under unbalanced data that customer churn prediction accuracy rate caused by the imbalance between classification is low, especially the low problem of high-value user's predictablity rate.
Description
Technical field
The present invention relates to technical field of information management, more particularly, to a kind of telecom client attrition prediction method, system
And electronic equipment.
Background technique
With the development of telecommunications industry, client selects the leeway of telecommunication product and telecommunications company increasing, telecom client
Churn management is particularly important in the more and more fierce telecommunications industry of competition.In the case where market tends to saturation, development is new
The cost of client is much larger than the cost for retaining existing customer.Therefore, predict that potential customer revenue is particularly important for company,
It is necessary to construct the model for being intended to find to have loss orientation client.
Currently, related network reason and non-network reason before the main churn prediction by analysis, building customer churn are pre-
Survey model, and the client that will be lost using the model prediction.Currently, there are many this kind of prediction model, such as machine learning algorithm packet
Include decision tree, neural network, SVM, random forest and based on their innovatory algorithm etc..Divided accordingly using these Algorithm constitutions
Class device can predict customer churn.
But neural network restrains the problem of slow and local minimum point etc., causes calculating error that can not decline.In addition, examining
Considering existing telecommunication data has the characteristics that data dimension is high, and random forest has, choosing low to data quality requirement
Select variable and sampling the features such as being random, can guarantee the accuracy of classification, and to exceptional value and noise have compared with
Good tolerance, therefore big more options random forests algorithm, to be carried out according to the behavior before telecom client off-network to customer churn
Prediction.
Although existing random forests algorithm is relatively high to the predictablity rate of customer churn, it is less prone to over-fitting.
But random forest, since customer churn ratio is very low, leads to sample centralized activity in the prediction application that telecom client is lost
Client is many more than customer revenue number, and such sample is nonequilibrium.And the meter of the Geordie radix of original random forest
It is calculated at last using all kinds of numbers, when causing to carry out the prediction of customer churn using random forest in this way, predicted value ratio
Compared with the more class of classification sample is biased to, preferably customer churn cannot be predicted.
Summary of the invention
In order to overcome the above problem or at least be partially solved the above problem, the present invention provides a kind of telecom client loss
Prediction technique, system and electronic equipment, effectively to solve under unbalanced data, customer churn caused by the imbalance between classification
Predictablity rate is low, especially the low problem of high-value user's predictablity rate.
In a first aspect, the present invention provides a kind of telecom client attrition prediction method, comprising: calculate in telecommunication network in netter
Family and the corresponding sample division proportion of off-network client, and based on described at netter family and the corresponding sample of off-network client
Division proportion determines the improvement Geordie radix of random forests algorithm；Based on the improvement Geordie radix, calculated using random forest
Method, prediction telecom client are lost；Wherein, the sample division proportion indicates, the client of the correspondence classification of selection in net state
Sample size accounts for the category in the ratio of net state client's total amount.
Further, at netter family and the corresponding sample division proportion of off-network client in the calculating telecommunication network
The step of before, the method also includes the values of life based on telecom client, are arranged different predictions for different telecom clients
Malfunction penalty term；Correspondingly, at netter family and the corresponding sample division proportion of off-network client in the calculating telecommunication network
The step of further comprise: malfunction penalty term, calculate described corresponding at netter family and off-network client in conjunction with the prediction
Sample division proportion.
Further, random forests algorithm is utilized described, before predicting the step of telecom client is lost, the method is also
It include: the threshold value for setting classification of voting in random forests algorithm；Correspondingly, described utilize random forests algorithm, prediction telecommunications visitor
The step of family is lost further comprises: according to the threshold value of the ballot classification, carrying out random forests algorithm operation, prediction telecommunications visitor
Family is lost.
Wherein, in the calculating telecommunication network the netter family and off-network client corresponding sample division proportion the step of
Further comprise: utilizing following quantity ratio, indicates the sample size of each partitioning site in random forests algorithm:
In formula, AR indicates the sample size of t partitioning site, and t indicates that the left sibling or right node in random forest tree, k indicate
For client in net state classification, value is in net or off-network, CtkIndicate sample size of the client in net state classification for k at t node,
CkIndicate client in the sample size that net state classification is k, CtIndicate t node sample size, λ1Indicate the first adjustment parameter；
Based on the sample size of each partitioning site, calculate described at netter family and the corresponding sample of off-network client
Division proportion is as follows:
In formula, ARP (k | t) indicates ratio of the sample AR value in node t in node t client in net state classification k, k
Value is in net or off-network, and AR (k=0 | t) and AR (k=1 | t) respectively indicate the client in node t in net and churn prediction
AR value.
Wherein, prediction error penalty term described in the combination calculates described corresponding at netter family and off-network client
The step of sample division proportion, further comprises: utilizing following quantity ratio, indicates each partitioning site base in random forests algorithm
In the sample size of client's value of life:
In formula, VAR indicates sample size of the t partitioning site based on client's value of life, and t indicates the left section in random forest tree
Point or right node, k indicate client in net state classification, and value is in net or off-network, VCtkIndicate that client is netted at t node
State classification is the sum of the sample value of life of k, VCkIndicate client in the sum of the sample value of life that net state classification is k, VCtTable
Show the sum of t node sample value of life, λ2Indicate the second adjustment parameter；
Based on the sample size of each partitioning site based on client's value of life, calculate described at netter family and off-network visitor
The corresponding sample division proportion in family is as follows:
In formula, VARP (k | t) indicate client at node t in ratio of the sample VAR value in node t of net state classification k,
K value be in net or off-network, VAR (k=0 | t) and VAR (k=1 | t) respectively indicate the client in node t net and client from
The VAR value of net.
Wherein, it is described based on described at netter family and the corresponding sample division proportion of off-network client, determine random gloomy
The step of improvement Geordie radix of woods algorithm, further comprises: being drawn with described at netter family and the corresponding sample of off-network client
Divide ratio, as the improvement Geordie radix.
Further, it in the sample size based on each partitioning site, calculates described at netter family and off-network visitor
After the step of corresponding sample division proportion in family, the method also includes: based on described at netter family and off-network client
Corresponding sample division proportion calculates the information gain of the sample size of each partitioning site.
Wherein, described to be based on the improvement Geordie radix, using random forests algorithm, predict the step of telecom client is lost
Further comprise: choosing the reckling in the information gain, is drawn as the division attribute of random forests algorithm, and based on described
Adhering to separately property creates each node of each tree in random forests algorithm, generates non-beta pruning tree；Based on all non-beta pruning trees,
Random forest is constituted, and utilizes the random forest partition testing collection, ballot classification is carried out, determines customer revenue.
Second aspect, the present invention provide a kind of telecom client attrition prediction system, comprising: setting module, for calculating electricity
At netter family and the corresponding sample division proportion of off-network client in communication network, and based on described at netter family and off-network client
Corresponding sample division proportion determines the improvement Geordie radix of random forests algorithm；Prediction module, for being changed based on described
Into Geordie radix, using random forests algorithm, predict that telecom client is lost；Wherein, the sample division proportion indicates, selection
Corresponding classification accounts for the category in the ratio of net state client's total amount in the sample size of the client of net state.
The third aspect, the present invention provide a kind of electronic equipment, comprising: at least one processor, is led at least one processor
Believe interface and bus；The memory, the processor and the communication interface complete mutual communication by the bus,
The communication interface is passed for the information of the electronic equipment and client between net state detection device and data setting equipment
It is defeated；The computer program that can be run on the processor is stored in the memory, the processor executes the calculating
When machine program, telecom client attrition prediction method as described above is realized.
A kind of telecom client attrition prediction method, system and electronic equipment provided by the invention are accounted for according to sample size is divided
Ratio inside this classification, the Geordie radix in the random forests algorithm of computed improved, to improve random forests algorithm
Telecom client attrition prediction, can effectively solve the problem that under unbalanced data, customer churn prediction caused by the imbalance between classification
Accuracy rate is low, especially the low problem of high-value user's predictablity rate.
Detailed description of the invention
Fig. 1 is a kind of flow chart of telecom client attrition prediction method of the embodiment of the present invention；
Fig. 2 is according to a kind of preferred process flow chart of telecom client attrition prediction method of the embodiment of the present invention；
Fig. 3 is a kind of structural schematic diagram of telecom client attrition prediction system of the embodiment of the present invention；
Fig. 4 is the structural block diagram of a kind of electronic equipment of the embodiment of the present invention.
Specific embodiment
To make the object, technical solutions and advantages of the present invention clearer, below in conjunction with attached in the embodiment of the present invention
Figure, is clearly and completely described the technical solution in the present invention, it is clear that described embodiment is one of the invention
Divide embodiment, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art are not making
Every other embodiment obtained, shall fall within the protection scope of the present invention under the premise of creative work.
For defect of the existing random forests algorithm in terms of unbalanced data, the embodiment of the present invention proposes a kind of improved
Random forests algorithm improves the telecom client attrition prediction of random forests algorithm that is, by computed improved Geordie radix.
As the one aspect of the embodiment of the present invention, the present embodiment provides a kind of telecom client attrition prediction method, references
Fig. 1 is a kind of flow chart of telecom client attrition prediction method of the embodiment of the present invention, comprising:
S1 is calculated in telecommunication network at netter family and the corresponding sample division proportion of off-network client, and based in net
Client and the corresponding sample division proportion of off-network client, determine the improvement Geordie radix of random forests algorithm；
S2, using random forests algorithm, predicts that telecom client is lost based on Geordie radix is improved；
Wherein, sample division proportion indicates, the correspondence classification of selection accounts for the category in the sample size of the client of net state and exists
The ratio of net state client's total amount.
It is to be understood that the present embodiment is solved when carrying out customer churn prediction using random forests algorithm, data are uneven
The problem of weighing apparatus.Specifically when calculating the Geordie radix of random forests algorithm, the ratio inside this classification is accounted for using sample size is divided,
Only in classification interior contrast, thus carry out the division between balancedunbalanced categories of datasets.Therefore, make different classes of in category division
Example it is unaffected, unbalanced situation is distributed between two classes without considering.
It is understood that above-mentioned classification is classifying in net state to all telecommunication users, such as it is included in net state
With off-network state, then correspondingly, corresponding classification in the client of net state is in the client of net state or the visitor of off-network state
Family.
Specifically in step sl, divided first according to determining random tree node division sample size using given calculation formula
It Ji Suan not be in the sample division proportion at netter family and the sample division proportion of off-network client.Then, using in netter family and off-network
The corresponding sample division proportion of client determines the Geordie radix of random forests algorithm, i.e. improvement Geordie radix.
Wherein, in one embodiment, it based at netter family and the corresponding sample division proportion of off-network client, determines
The step of improvement Geordie radix of random forests algorithm, further comprises: at netter family and the corresponding sample of off-network client
Division proportion, as improvement Geordie radix.
Wherein, in another embodiment, it calculates in telecommunication network at netter family and the corresponding sample of off-network client
The step of division proportion, further comprises:
Firstly, indicating the sample size of each partitioning site in random forests algorithm using following quantity ratio:
In formula, AR indicates the sample size of t partitioning site, and t indicates that the left sibling or right node in random forest tree, k indicate
For client in net state classification, value is in net or off-network, CtkIndicate sample size of the client in net state classification for k at t node,
CkIndicate client in the sample size that net state classification is k, CtIndicate t node sample size, λ1Indicate the first adjustment parameter；
Then, the sample size based on each partitioning site is calculated and is drawn at netter family and the corresponding sample of off-network client
Divide ratio as follows:
In formula, ARP (k | t) indicates ratio of the sample AR value in node t in node t client in net state classification k, k
Value is in net or off-network, and AR (k=0 | t) and AR (k=1 | t) respectively indicate the client in node t in net and churn prediction
AR value.
It is understood that the ratio inside this classification is accounted for using sample size is divided, only in class when calculating Geordie radix
Other interior contrast.First with above-mentioned quantity ratio (Amount Ratio) AR, the sample size of each partitioning site is indicated.Wherein λ
Effect be exactly to be finely adjusted in conjunction with the different classes of data volume of same intra-node, to improve the anti-of same category interior contrast
Making an uproar property.
Division (in net and off-network) is exactly the distribution for changing unbalanced dataset between balancing classification, so that algorithm will not be because
And this is by too big influence, that is, improves the accuracy rate that off-network client divides.
Each information gain Δ is calculated in new data, need to calculate each AR value ratio shared by each sample value
Example (Amount Ratio Poportion) ARP, is calculated with specific reference to above-mentioned ARP calculation formula.
If ARP value is very high, in prediction classification, ratio shared by this kind is also high, and opposite AR is also big, then
It is also big that this kind of clients is predicted to be this kind of shared ratios.It therefore no matter is more in the ratio of net and off-network client
Few, impurity level measurement under this rule is not influenced by two classification client's ratios are unbalanced, is overcome traditional random gloomy
This disadvantage of woods.
Specifically in step s 2, on the basis of determining the improvement Geordie radix of random forests algorithm according to above-mentioned steps,
Using the improvement Geordie radix, the operation of random forests algorithm is carried out to the other client of telecommunications to be predicted, is realized to the telecommunications
The prediction of classification loss user.
A kind of telecom client attrition prediction method provided in an embodiment of the present invention accounts for inside this classification according to sample size is divided
Ratio, the Geordie radix in the random forests algorithm of computed improved, to improve the telecom client of random forests algorithm
Attrition prediction can effectively solve the problem that under unbalanced data that customer churn prediction accuracy rate caused by the imbalance between classification is low, especially
It is the low problem of high-value user's predictablity rate.
Further, on the basis of the above embodiments, distinguish in calculating telecommunication network at netter family and off-network client
Before the step of corresponding sample division proportion, this method further include: the value of life based on telecom client, for different telecommunications visitors
Different prediction error penalty terms is arranged in family；
Correspondingly, calculate in telecommunication network the netter family and off-network client corresponding sample division proportion the step of into
One step includes: to calculate in conjunction with prediction error penalty term at netter family and the corresponding sample division proportion of off-network client.
It is understood that considering that tradition, when being predicted, has ignored the life valence of client using random forest method
Value does not consider the accuracy of high value customer as far as possible when carrying out customer traffic prediction especially, has been easy to cause off-network tendency
High value customer sorted out by mistake, make its loss, and the loss of high value customer can cause huge loss to telecommunications company,
The present embodiment predicts that error assigns different penalty terms to different clients, and punishment item size is determined by the value of life of client.
Later, it is based on innovatory algorithm same as above-described embodiment, calculates the sample value of each division points, with based on client
The quantity of value of life balances the division in net and off-network client than indicating after client's value of life is added.
Wherein, in one embodiment, in conjunction with prediction error penalty term, calculating respectively corresponds at netter family and off-network client
Sample division proportion the step of further comprise:
Firstly, indicating that each partitioning site is based on client's value of life in random forests algorithm using following quantity ratio
Sample size:
In formula, VAR indicates sample size of the t partitioning site based on client's value of life, and t indicates the left section in random forest tree
Point or right node, k indicate client in net state classification, and value is in net or off-network, VCtkIndicate that client is netted at t node
State classification is the sum of the sample value of life of k, VCkIndicate client in the sum of the sample value of life that net state classification is k, VCtTable
Show the sum of t node sample value of life, λ2Indicate the second adjustment parameter；
Secondly, the sample size based on each partitioning site based on client's value of life, calculates at netter family and off-network client
Corresponding sample division proportion is as follows:
In formula, VARP (k | t) indicate client at node t in ratio of the sample VAR value in node t of net state classification k,
K value be in net or off-network, VAR (k=0 | t) and VAR (k=1 | t) respectively indicate the client in node t net and client from
The VAR value of net.
It is understood that being put down only improving unbalanced data to the improvement of random forest according to above-described embodiment
The problem of classification that weighs, but there is no the values of life for considering client.The value of each client is different, loss value
High client loses bigger caused by telecommunications company.
The present embodiment calculates the sample value of each division points also according to the improvement of the above method, is based on client with above-mentioned
The quantity ratio VAR of value of life is indicated after client's value of life is added, and balances the division at netter family and off-network client.And it is and every
A VAR value ratio shared by each node, also according to the present embodiment at netter family and the corresponding sample of off-network client
Division proportion formula is calculated.
A kind of telecom client attrition prediction method provided in an embodiment of the present invention is accounted in this classification using division sample size
On the basis of the ratio in portion calculates Geordie radix, propose that the node purity based on client's value of life divides, by the life of client
Value is integrated into improved gini index, can be improved off-network client's predictablity rate, and especially high value customer prediction is quasi-
True rate, thus to avoid the loss of high value customer from providing instruction.
Further, it in the sample size based on each partitioning site, calculates corresponding at netter family and off-network client
After the step of sample division proportion, this method further include: based at netter family and the corresponding sample division of off-network client
Ratio calculates the information gain of the sample size of each partitioning site.
It is understood that for the calculation method of above-mentioned improvement Geordie radix, in improved random forest calculating process
In, it further include the sample size VAR of sample size AR or each partitioning site based on client's value of life to each partitioning site
The calculating of corresponding information gain.
Wherein, particularly for the sample size AR of each partitioning site, it is as follows to calculate information gain:
In formula, ΔARThe information gain of the sample size AR of expression partitioning site, and ARP (k | t) it indicates in node t client in net
Ratio of the sample AR value of status categories k in node t, k value are AR (k=0 | t=l) and AR (k=1 in net or off-network
| t=l) respectively indicate the client in node l in the AR value of net and churn prediction, AR (k=0 | t=r) and AR (k=1 | t=r)
The client in node r is respectively indicated in the AR value of net and churn prediction.
Wherein, the sample size VAR particularly for each partitioning site based on client's value of life calculates information gain such as
Under:
The division between unbalanced data classification is balanced by above improvement, and the prediction for improving high-value user is quasi-
True rate.
It is wherein optional, based on improving Geordie radix, using random forests algorithm, predict the step of telecom client is lost into
One step includes:
The reckling in information gain is chosen, as the division attribute of random forests algorithm, and based on attribute is divided, is created
Each node of each tree in random forests algorithm, generates non-beta pruning tree；
Based on all non-beta pruning trees, random forest is constituted, and utilizes random forest partition testing collection, carries out ballot classification,
Determine customer revenue.
It is understood that according to above-described embodiment, for each AR value or VAR value, it being calculated and respectively corresponds
Information gain.For all AR values or VAR value, range of information yield value can get according to above-mentioned calculating, then choosing should
The smallest information gain value in series of values, the division attribute as random forests algorithm.Later, under this division attribute, wound
The each node for building each tree in random forests algorithm generates non-beta pruning tree.
Non- beta pruning tree is generated to every group of training sample according to above-mentioned processing step, then carries out K training producible K and non-cuts
Branch tree.Random forest is constituted using K non-beta pruning tree of generation, and using the random forest partition testing collection constituted, with setting
Threshold value compare, carry out ballot classification, and determine customer revenue according to classification results.
Further, on the basis of the above embodiments, random forests algorithm, the step that prediction telecom client is lost are being utilized
Before rapid, this method further include: the threshold value for classification of voting in setting random forests algorithm；
Correspondingly, predicting that the step of telecom client is lost further comprises using random forests algorithm: being classified according to ballot
Threshold value, carry out random forests algorithm operation, prediction telecom client be lost.
It is understood that under normal circumstances, telecommunications company concerns that off-network client is predicted correctly, and off-network client
Sorted out by mistake big to the loss of telecommunications company's bring by mistake classification more than at netter family；And original random forest side
Method, is equivalent to the classifier that threshold value is spanning tree half, and threshold value expression is predicted the threshold that classification should be classified as certain class；
So under this threshold, customer revenue is easily classified as at netter family.
Consider such case, the present embodiment after obtaining the random forest obtained by beta pruning tree according to above-described embodiment,
When carrying out ballot classification, appropriate threshold value is manually set, determines which kind of prediction classification should be classified as.I.e. by the way that threshold is manually set
The sample is divided into loss visitor if the number that a certain sample is divided into customer revenue tree to be greater than to the threshold value of setting by value
Family sample.
Due to customer revenue ratio very little, this thought, which is equivalent to, reduces the threshold for being classified as customer revenue, therefore can incite somebody to action
More customer revenues are correctly sorted out.
A kind of telecom client attrition prediction method provided in an embodiment of the present invention, by changing the value of life introduction of client
Into Geordie radix, and random forest vote classify when set appropriate threshold value, improve unbalanced data to customer churn
The influence of predictablity rate, and improve the customer churn prediction accuracy rate of high-value user.
To further illustrate technical solution of the present invention, following preferred process flow is provided, but not to guarantor of the invention
Shield range is limited.
With reference to Fig. 2, to be somebody's turn to do according to a kind of preferred process flow chart of telecom client attrition prediction method of the embodiment of the present invention
Input is training sample (X in preferred flow1,Y1), (X2,Y2) ..., (Xn,Yn), it exports as using ballot mode, according to threshold value T
The test set classification of decision.Specific process flow is as follows:
Firstly, sample is divided into n parts, using cross validation, the optimum value of attribute m is randomly choosed on each node；
Secondly, sequentially generating k decision tree；
Again, from the training set of n sample, repeat to extract n sample as training set；And randomly choose m category
Property, node split is carried out, using the calculation formula of the information gain of the sample size VAR based on client's value of life, calculates information
Gain；And select information gain the smallest as attribute is divided, each node of each tree is successively created, non-beta pruning tree is generated；
Finally, constituting random forest using the k tree generated, thrown using random forest partition testing collection with threshold comparison
Ticket classification.
As the other side of the embodiment of the present invention, the present embodiment provides a kind of telecom client stream according to above-described embodiment
Forecasting system is lost, is a kind of structural schematic diagram of telecom client attrition prediction system of the embodiment of the present invention with reference to Fig. 3.It include: to set
Cover half block 1 and prediction module 2.
Wherein, setting module 1 is divided for calculating in telecommunication network at netter family and the corresponding sample of off-network client
Ratio, and based at netter family and the corresponding sample division proportion of off-network client, determine the improvement base of random forests algorithm
Thessaloniki number；Prediction module 2 is used to, using random forests algorithm, predict that telecom client is lost based on Geordie radix is improved；Wherein,
Sample division proportion indicates that the correspondence classification of selection accounts for the category in net state client's total amount in the sample size of the client of net state
Ratio.
It is understood that the telecom client attrition prediction system in the embodiment of the present invention is in the above embodiments
Realize the prediction being lost to telecom client.Therefore, the description in telecom client attrition prediction method in the above embodiments
And definition, it can be used for the understanding of each execution module in the embodiment of the present invention.
It is understood that can be by hardware processor (hardware processor) come real in the embodiment of the present invention
Existing related function module.
A kind of telecom client attrition prediction system provided in an embodiment of the present invention, by the way that corresponding functional module is arranged,
On the basis of accounting for the calculating Geordie radix of the ratio inside this classification using division sample size, the section based on client's value of life is proposed
Point purity divides, and the value of life of client is integrated into improved gini index, can be improved off-network client's predictablity rate,
Especially high value customer predictablity rate, thus to avoid the loss of high value customer from providing instruction.
As the another aspect of the embodiment of the present invention, the present embodiment provides a kind of electronic equipment according to above-described embodiment,
Be the structural block diagram of a kind of electronic equipment of the embodiment of the present invention with reference to Fig. 4, comprising: at least one processor 401, at least one
Processor 402, communication interface 403 and bus 404.
Wherein, memory 401, processor 402 and communication interface 403 complete mutual communication by bus 404, communicate
Interface 403 is transmitted for the information of the electronic equipment and client between net state detection device and data setting equipment；Storage
The computer program that can be run on the processor 402 is stored in device 401, it is real when processor 402 executes the computer program
The now telecom client attrition prediction method as described in above-described embodiment.
It is to be understood that including at least memory 401, processor 402,403 and of communication interface in the electronic equipment
Bus 404, and memory 401, processor 402 and communication interface 403 form mutual communication connection by bus 404,
And achievable mutual communication.
Communication interface 403 realizes that the electronic equipment and client are logical between net state detection device and data setting equipment
Letter connection, and achievable mutual information transmission, are such as realized by communication interface 403 and are joined to client in net state and user preset
Several acquisitions etc..
When electronic equipment is run, processor 402 calls the program instruction in memory 401, real to execute above-mentioned each method
Apply method provided by example, for example, calculate in telecommunication network and divide at netter family and the corresponding sample of off-network client
Ratio, and changing for random forests algorithm is determined at netter family and the corresponding sample division proportion of off-network client based on described
Into Geordie radix；And the value of life based on telecom client, different prediction error penalty terms is set for different telecom clients
Deng.
In another embodiment of the present invention, a kind of non-transient computer readable storage medium, the non-transient calculating are provided
Machine readable storage medium storing program for executing stores computer instruction, and the computer instruction executes the computer as described in above-described embodiment
Telecom client attrition prediction method.
It is to be understood that the logical order in above-mentioned memory 401 can be realized by way of SFU software functional unit
And when sold or used as an independent product, it can store in a computer readable storage medium.Alternatively, on realizing
Stating all or part of the steps of embodiment of the method, this can be accomplished by hardware associated with program instructions, and program above-mentioned can deposit
It is stored in a computer-readable storage medium, which when being executed, executes step including the steps of the foregoing method embodiments；And it is preceding
The storage medium stated includes: various Jie that can store program code such as USB flash disk, mobile hard disk, ROM, RAM, magnetic or disk
Matter.
The embodiment of electronic equipment described above is only schematical, wherein unit as illustrated by the separation member
It may or may not be physically separated, both can be located in one place, or may be distributed over heterogeneous networks
On unit.Some or all of the modules therein can be selected to achieve the purpose of the solution of this embodiment according to actual needs.This
Field those of ordinary skill is without paying creative labor, it can understands and implements.
By the description of embodiment of above, those skilled in the art is it will be clearly understood that each embodiment can borrow
Help software that the mode of required general hardware platform is added to realize, naturally it is also possible to pass through hardware.Based on this understanding, above-mentioned
Substantially the part that contributes to existing technology can be embodied in the form of software products technical solution in other words, the meter
Calculation machine software product may be stored in a computer readable storage medium, such as USB flash disk, mobile hard disk, ROM, RAM, magnetic disk or light
Disk etc., including some instructions, with so that a computer equipment (such as personal computer, server or network equipment etc.)
Execute method described in certain parts of above-mentioned each method embodiment or embodiment of the method.
A kind of electronic equipment provided in an embodiment of the present invention and a kind of non-transient computer readable storage medium are drawn using
Point sample size accounts on the basis of the ratio inside this classification calculates Geordie radix, proposes the node purity based on client's value of life
It divides, the value of life of client is integrated into improved gini index, can be improved off-network client's predictablity rate, especially
High value customer predictablity rate, thus to avoid the loss of high value customer from providing instruction.
In addition, those skilled in the art are it should be understood that in application documents of the invention, term " includes ",
"comprising" or any other variant thereof is intended to cover non-exclusive inclusion so that include a series of elements process,
Method, article or equipment not only include those elements, but also including other elements that are not explicitly listed, or are also wrapped
It includes as elements inherent to such a process, method, article, or device.In the absence of more restrictions, by sentence " including
One ... " limit element, it is not excluded that there is also another in the process, method, article or apparatus that includes the element
Outer identical element.
In specification of the invention, numerous specific details are set forth.It should be understood, however, that the embodiment of the present invention can
To practice without these specific details.In some instances, well known method, structure and skill is not been shown in detail
Art, so as not to obscure the understanding of this specification.Similarly, it should be understood that disclose in order to simplify the present invention and helps to understand respectively
One or more of a inventive aspect, in the above description of the exemplary embodiment of the present invention, each spy of the invention
Sign is grouped together into a single embodiment, figure, or description thereof sometimes.
However, the disclosed method should not be interpreted as reflecting the following intention: i.e. the claimed invention requirement
Features more more than feature expressly recited in each claim.More precisely, as claims are reflected
Like that, inventive aspect is all features less than single embodiment disclosed above.Therefore, it then follows the power of specific embodiment
Thus sharp claim is expressly incorporated in the specific embodiment, wherein each claim itself is as independent reality of the invention
Apply example.
Finally, it should be noted that the above embodiments are merely illustrative of the technical solutions of the present invention, rather than its limitations；Although
Present invention has been described in detail with reference to the aforementioned embodiments, and those skilled in the art is it is understood that it still can be right
Technical solution documented by foregoing embodiments is modified or equivalent replacement of some of the technical features；And this
It modifies or replaces, the spirit and model of technical solution of various embodiments of the present invention that it does not separate the essence of the corresponding technical solution
It encloses.
Claims (10)
1. a kind of telecom client attrition prediction method characterized by comprising
It calculates in telecommunication network at netter family and the corresponding sample division proportion of off-network client, and based on described at netter family
Sample division proportion corresponding with off-network client determines the improvement Geordie radix of random forests algorithm；
Based on the improvement Geordie radix, using random forests algorithm, predict that telecom client is lost；
Wherein, the sample division proportion indicates, the correspondence classification of selection accounts for the category in the sample size of the client of net state and exists
The ratio of net state client's total amount.
2. the method according to claim 1, wherein at netter family and off-network visitor in the calculating telecommunication network
Before the step of corresponding sample division proportion in family, further includes:
Different prediction error penalty terms is arranged for different telecom clients for value of life based on telecom client；
It is further wrapped in the calculating telecommunication network in the step of netter family and off-network client corresponding sample division proportion
It includes:
In conjunction with prediction error penalty term, calculate described at netter family and the corresponding sample division proportion of off-network client.
3. predicting telecom client the method according to claim 1, wherein utilizing random forests algorithm described
Before the step of loss, further includes:
Set the threshold value for classification of voting in random forests algorithm；
Correspondingly, the step of described utilize random forests algorithm, and prediction telecom client is lost further comprises:
According to the threshold value of the ballot classification, random forests algorithm operation is carried out, prediction telecom client is lost.
4. the method according to claim 1, wherein at netter family and off-network client in the calculating telecommunication network
The step of corresponding sample division proportion, further comprises:
Using following quantity ratio, the sample size of each partitioning site in random forests algorithm is indicated:
In formula, AR indicates the sample size of t partitioning site, and t indicates that the left sibling or right node in random forest tree, k indicate client
In net state classification, value is in net or off-network, CtkIndicate that client is in the sample size that net state classification is k, C at t nodekTable
Show client in the sample size that net state classification is k, CtIndicate t node sample size, λ1Indicate the first adjustment parameter；
Based on the sample size of each partitioning site, calculate described at netter family and the corresponding sample division of off-network client
Ratio is as follows:
In formula, ARP (k | t) indicates ratio of the sample AR value in node t in node t client in net state classification k, k value
For in net or off-network, AR (k=0 | t) and AR (k=1 | t) client in node t is respectively indicated in the AR of net and churn prediction
Value.
5. according to the method described in claim 2, it is characterized in that, error penalty term is predicted described in the combination, described in calculating
Further comprise in the step of netter family and off-network client corresponding sample division proportion:
Using following quantity ratio, sample size of each partitioning site based on client's value of life in random forests algorithm is indicated:
In formula, VAR indicate sample size of the t partitioning site based on client's value of life, t indicate random forest tree in left sibling or
Right node, k indicate client in net state classification, and value is in net or off-network, VCtkIndicate that client is in net state class at t node
Not Wei k the sum of sample value of life, VCkIndicate client in the sum of the sample value of life that net state classification is k, VCtIndicate t
The sum of node sample value of life, λ2Indicate the second adjustment parameter；
Based on the sample size of each partitioning site based on client's value of life, calculate described at netter family and off-network client point
Not corresponding sample division proportion is as follows:
In formula, VARP (k | t) indicates client at node t in ratio of the sample VAR value in node t of net state classification k, and k takes
For value in net or off-network, VAR (k=0 | t) and VAR (k=1 | t) respectively indicate the client in node t in net and churn prediction
VAR value.
6. according to the method described in claim 4, it is characterized in that, described based on described right respectively at netter family and off-network client
The sample division proportion answered, the step of determining the improvement Geordie radix of random forests algorithm, further comprise:
With described at netter family and the corresponding sample division proportion of off-network client, as the improvement Geordie radix.
7. according to the method described in claim 6, it is characterized in that, in the sample size based on each partitioning site,
Calculate it is described the netter family and off-network client corresponding sample division proportion the step of after, further includes:
Based on described at netter family and the corresponding sample division proportion of off-network client, the sample of each partitioning site is calculated
The information gain of this amount.
8. the method according to the description of claim 7 is characterized in that it is described be based on the improvements Geordie radix, using at random it is gloomy
The step of woods algorithm, prediction telecom client is lost, further comprises:
The reckling in the information gain is chosen, as the division attribute of random forests algorithm, and is based on the division attribute,
The each node for creating each tree in random forests algorithm, generates non-beta pruning tree；
Based on all non-beta pruning trees, random forest is constituted, and utilizes the random forest partition testing collection, carries out ballot point
Class determines customer revenue.
9. a kind of telecom client attrition prediction system characterized by comprising
Setting module, for calculating in telecommunication network at netter family and the corresponding sample division proportion of off-network client, and base
In described at netter family and the corresponding sample division proportion of off-network client, the improvement Geordie base of random forests algorithm is determined
Number；
Prediction module, using random forests algorithm, predicts that telecom client is lost for being based on the improvement Geordie radix；
Wherein, the sample division proportion indicates, the correspondence classification of selection accounts for the category in the sample size of the client of net state and exists
The ratio of net state client's total amount.
10. a kind of electronic equipment characterized by comprising at least one processor, at least one processor, communication interface and
Bus；
The memory, the processor and the communication interface complete mutual communication, the communication by the bus
Interface is transmitted for the information of the electronic equipment and client between net state detection device and data setting equipment；
The computer program that can be run on the processor is stored in the memory, the processor executes the calculating
When machine program, the method as described in any in claim 1 to 8 is realized.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810871287.3A CN109190796B (en) | 2018-08-02 | 2018-08-02 | Telecommunication customer loss prediction method, system and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810871287.3A CN109190796B (en) | 2018-08-02 | 2018-08-02 | Telecommunication customer loss prediction method, system and electronic equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109190796A true CN109190796A (en) | 2019-01-11 |
| CN109190796B CN109190796B (en) | 2021-03-02 |
Family
ID=64920571
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810871287.3A Active CN109190796B (en) | 2018-08-02 | 2018-08-02 | Telecommunication customer loss prediction method, system and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109190796B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111259975A (en) * | 2020-01-21 | 2020-06-09 | 支付宝(杭州)信息技术有限公司 | Method and device for generating classifier and method and device for classifying text |
| CN112767125A (en) * | 2021-01-15 | 2021-05-07 | 上海琢学科技有限公司 | Customer loss prediction method, device and storage medium |
| CN113240518A (en) * | 2021-07-12 | 2021-08-10 | 广州思迈特软件有限公司 | Bank-to-public customer loss prediction method based on machine learning |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2006190047A (en) * | 2005-01-05 | 2006-07-20 | Nippon Signal Co Ltd:The | Vehicle selection system |
| CN106022505A (en) * | 2016-04-28 | 2016-10-12 | 华为技术有限公司 | Method and device of predicting user off-grid |
| CN107818376A (en) * | 2016-09-13 | 2018-03-20 | 中国电信股份有限公司 | Customer loss Forecasting Methodology and device |
| CN108280652A (en) * | 2016-12-31 | 2018-07-13 | 中国移动通信集团辽宁有限公司 | The analysis method and device of user satisfaction |
-
2018
- 2018-08-02 CN CN201810871287.3A patent/CN109190796B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2006190047A (en) * | 2005-01-05 | 2006-07-20 | Nippon Signal Co Ltd:The | Vehicle selection system |
| CN106022505A (en) * | 2016-04-28 | 2016-10-12 | 华为技术有限公司 | Method and device of predicting user off-grid |
| CN107818376A (en) * | 2016-09-13 | 2018-03-20 | 中国电信股份有限公司 | Customer loss Forecasting Methodology and device |
| CN108280652A (en) * | 2016-12-31 | 2018-07-13 | 中国移动通信集团辽宁有限公司 | The analysis method and device of user satisfaction |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111259975A (en) * | 2020-01-21 | 2020-06-09 | 支付宝(杭州)信息技术有限公司 | Method and device for generating classifier and method and device for classifying text |
| CN111259975B (en) * | 2020-01-21 | 2022-07-22 | 支付宝(杭州)信息技术有限公司 | Method and device for generating classifier and method and device for classifying text |
| CN112767125A (en) * | 2021-01-15 | 2021-05-07 | 上海琢学科技有限公司 | Customer loss prediction method, device and storage medium |
| CN113240518A (en) * | 2021-07-12 | 2021-08-10 | 广州思迈特软件有限公司 | Bank-to-public customer loss prediction method based on machine learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109190796B (en) | 2021-03-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110119413A (en) | The method and apparatus of data fusion | |
| CN113935434A (en) | Data analysis processing system and automatic modeling method | |
| CN109190796A (en) | A kind of telecom client attrition prediction method, system and electronic equipment | |
| CN110088749A (en) | Automated ontology generates | |
| CN109242361A (en) | A kind of fire-fighting methods of risk assessment, device and terminal device | |
| CN108038052A (en) | Automatic test management method, device, terminal device and storage medium | |
| CN112491854B (en) | Multi-azimuth security intrusion detection method and system based on FCNN | |
| CN107590196A (en) | Earthquake emergency information screening and evaluating system and system in a kind of social networks | |
| CN106803039B (en) | A kind of homologous determination method and device of malicious file | |
| CN108564423A (en) | Malice occupy-place recognition methods, system, equipment and the storage medium of ticketing service order | |
| CN109558384A (en) | Log classification method, device, electronic equipment and storage medium | |
| CN110222733A (en) | The high-precision multistage neural-network classification method of one kind and system | |
| CN115147092A (en) | Resource approval method and training method and device of random forest model | |
| CN107256461B (en) | Charging facility construction address evaluation method and system | |
| EP3185184A1 (en) | The method for analyzing a set of billing data in neural networks | |
| CN111275485A (en) | Power grid customer grade division method and system based on big data analysis, computer equipment and storage medium | |
| CN111126627A (en) | Model training system based on separation degree index | |
| CN114139931A (en) | Enterprise data evaluation method and device, computer equipment and storage medium | |
| CN107871055A (en) | A kind of data analysing method and device | |
| Śniegula et al. | Study of machine learning methods for customer churn prediction in telecommunication company | |
| CN116402546A (en) | Store risk attribution method and device, equipment, medium and product thereof | |
| CN110377809A (en) | The resource acquisition qualification generation method and relevant device of pre-set user | |
| CN112100165B (en) | Traffic data processing method, system, equipment and medium based on quality assessment | |
| CN114676253A (en) | Metadata hierarchical classification method based on machine learning algorithm | |
| CN115310865A (en) | Product full-quality life cycle quality control platform based on cosmetic detection technology |
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 |