CN115456260A - Customer service telephone traffic prediction method - Google Patents
Customer service telephone traffic prediction method Download PDFInfo
- Publication number
- CN115456260A CN115456260A CN202211051280.XA CN202211051280A CN115456260A CN 115456260 A CN115456260 A CN 115456260A CN 202211051280 A CN202211051280 A CN 202211051280A CN 115456260 A CN115456260 A CN 115456260A
- Authority
- CN
- China
- Prior art keywords
- model
- data
- customer service
- prediction
- predicting
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 30
- 238000007781 pre-processing Methods 0.000 claims abstract description 6
- 238000013499 data model Methods 0.000 claims abstract description 4
- 238000012545 processing Methods 0.000 claims description 5
- 238000012549 training Methods 0.000 claims description 5
- 230000002159 abnormal effect Effects 0.000 claims description 3
- 238000005096 rolling process Methods 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims description 2
- 238000009499 grossing Methods 0.000 claims description 2
- 230000003203 everyday effect Effects 0.000 claims 1
- 230000001932 seasonal effect Effects 0.000 claims 1
- 238000004422 calculation algorithm Methods 0.000 abstract description 8
- 230000000694 effects Effects 0.000 abstract description 4
- 230000007547 defect Effects 0.000 abstract description 2
- 238000012360 testing method Methods 0.000 description 4
- 238000010801 machine learning Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 230000004927 fusion Effects 0.000 description 2
- 238000007726 management method Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 230000001373 regressive effect Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
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
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/21—Design, administration or maintenance of databases
- G06F16/215—Improving data quality; Data cleansing, e.g. de-duplication, removing invalid entries or correcting typographical errors
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Tourism & Hospitality (AREA)
- Health & Medical Sciences (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Computing Systems (AREA)
- Artificial Intelligence (AREA)
- Strategic Management (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Marketing (AREA)
- Databases & Information Systems (AREA)
- General Business, Economics & Management (AREA)
- Quality & Reliability (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Development Economics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a method for predicting the traffic of customer service telephone traffic, which mainly comprises the following steps: 1) data preprocessing, 2) data model building and 3) building a linear programming. According to the scheme, a linear programming is adopted, and each model is given weight, so that the influence of each model on final prediction is different, and a good-bad complementation effect is achieved. The method overcomes the defects of low accuracy and low robustness of a single model, adopts a method similar to an integrated algorithm and a lifting algorithm, and simultaneously considers factors such as holidays and festivals, so that the accuracy of a prediction result is improved by one grade.
Description
Technical Field
The invention relates to the technical field of electronic commerce services, in particular to a prediction method for predicting input telephone traffic of future days according to the input telephone traffic of existing customer services.
Background
With the increasing expansion of the scale of call centers and service centers, the requirements of operation management are continuously improved, how to achieve the target of customer service level under the existing manpower condition, reasonably arrange the manpower, and optimize the field management becomes a huge challenge for the shift arrangement engineer. Time series predictive analysis uses the characteristics of an event at a time in the past to predict the characteristics of the event at a future time. The time series model is dependent on the sequence of events, and the results generated by inputting the time series model after the sequence of values with the same size is changed are different. Time series can be divided into stationary sequences, i.e., sequences in which there is some periodicity, and variations and means of seasonality and trends do not change over time, and non-stationary sequences.
The traditional prediction methods are divided into two types, one is a moving average method, an exponential average method and the like; the modeling steps based on the method are that firstly, stationarity detection needs to be carried out on an observed value sequence, and if the observed value sequence is unstable, differential operation is carried out on the observed value sequence until the data after the difference is stable; after the data are stable, carrying out white noise detection on the data, wherein the white noise is a random stable sequence of zero mean constant variance; if the sequence is a stable non-white noise sequence, calculating ACF (autocorrelation coefficient) and PACF (partial autocorrelation coefficient), carrying out model identification such as ARMA (auto regressive moving average), determining model parameters of the identified model, and finally applying prediction and carrying out error analysis.
Modern time series prediction methods favor machine learning as well as deep learning methods. For machine learning methods, xgboost, random forest and support vector machines are available. The key point of the data mining method is characteristic engineering, and different from other mining tasks, the characteristic engineering of the time series uses a sliding window, namely, data indexes such as maximum values, minimum values, mean values, variances and the like in the sliding window are calculated to be used as new characteristics. For deep learning methods, the lstm neural network is at best suited to address such methods.
The method has the advantages that the time sequence data has high randomness and is influenced by factors such as accuracy, periodicity, seasonality and the like, and the traditional method has high prediction difficulty and even has serious problems such as hysteresis.
Disclosure of Invention
The invention aims to overcome the problems in the prior art and provide a customer service telephone traffic prediction method.
In order to achieve the purpose, the technical scheme of the invention is as follows:
the method for predicting the customer service telephone traffic is characterized by comprising the following steps
1) Data pre-processing
Filling missing values, and not deleting abnormal values;
2) Data model building
Respectively establishing four basic models of lstm, holt-winter, xgboost and prophet, wherein the data processing of each model is consistent, but the input data formats of the models are different, and the data preprocessing relates to missing value filling and feature construction;
3) Establishing a linear program
The objective is to predict the data of 45 days in the future according to the historical data and extract the predicted data of the next month, the requirement for the model is that the predicted value must be in the interval of 98.4% and 108.4% of the actual value, and the following linear programming problem is established according to the prediction result of the basic model:
knowing that X is the historical data of the input, f i Is each basic model, Y is the actual traffic data, α i The weight of each of the basic models is represented,
min b＝|∑α i f i (X)-Y|
0.984Y≤b≤1.804Y
the linear programming is established to obtain alpha i Then by sigma alpha i f i (X) as a final model to predict future data.
In the step 2), the lstm model and the xgboost model adopt rolling prediction and are trained on gpu.
Wherein in step 2), the Prophet model adds holidays as factor parameters.
The invention has the beneficial effects that:
when the xgboost and lstm models are used, a rolling prediction mode is adopted, for example, the traffic volume of the future half day is predicted, the newly predicted traffic volume is added into the previous training set again, then training is carried out, and the future half day is predicted, so that the effect of predicting the traffic volume of the future week is better than that of directly predicting the traffic volume of the future week. In addition, the time series prediction often has the problem of a lag term, the reason for generating the lag term is that the sequence has autocorrelation, and a difference operation needs to be performed in the processing process, namely, the difference value between the current moment and the previous moment is used as a regression target.
Finally, a transverse and longitudinal prediction mode is adopted, wherein the transverse mode is model training according to the normal traffic sampled every quarter clock, the longitudinal mode is sampling every other day, such as 12 the first day and 12 the second day, 30 \8230, the same time of a certain day in the future is predicted, and then the transverse and longitudinal prediction results are averaged, so that the prediction result is better than the unidirectional prediction result.
According to the scheme, a linear programming is adopted, and each model is given with weight, so that the influence of each model on final prediction is different, and a role of complementing quality is played.
The method overcomes the defects of low accuracy and low robustness of a single model, adopts a method similar to an integrated algorithm and a lifting algorithm, and simultaneously considers factors such as holidays and festivals, so that the accuracy of a prediction result is improved by one grade.
Drawings
FIG. 1 is a prediction flow chart according to the present invention.
Fig. 2 is a diagram of an example of traffic data in a month of a year.
FIG. 3 is a graph of time series stability analysis.
FIG. 4 shows the holt-winter model time series prediction.
FIG. 5 is a time series prediction of the lstm model.
FIG. 6 is an xgboost model time series prediction.
FIG. 7 shows prophet model time series prediction
FIG. 8 is the final prediction for 4 models fused together
Detailed Description
The invention will be further illustrated with reference to the following specific examples.
The scheme predicts the input telephone traffic of the future days according to the past customer service input telephone traffic, and belongs to the category of time series modeling
Step 1, the telephone traffic data is not stable through DF-TEST TEST, and the data of all the years is stable basically, so that the data set is very suitable for time series modeling to carry out prediction analysis.
And 2, preprocessing data, wherein model data has a plurality of missing values and influences a prediction result, so that the missing values need to be filled, and in addition, abnormal values are important and cannot be deleted.
Step 3, establishing a data model, wherein four basic models of lstm, holt-winter, xgboost and prophet are respectively established, the data processing of each model is consistent, but the input data of the models are different, the data are respectively integrated into corresponding input, and the prediction result of each model is obtained and recorded as
Step 4, a linear programming problem is established, the aim is to predict data of 45 days in the future according to historical data and extract predicted data of the next month, the requirement of the model is that the predicted value must be within the interval of 98.4% and 108.4% of the actual value, so the following linear programming problem can be established according to the prediction result of the basic model:
knowing that X is the historical data of the input, f i Is each basic model, Y is actual traffic data, α i The weight of each of the basic models is represented,
min b＝|∑α i f i (X)-Y|
0.984Y≤b≤1.804Y
the linear programming problem is established to obtain alpha i Then by sigma alpha i f i (X) predicting future data as a final model, the results of the hybrid model improving accuracy by at least 10% over the single model.
FIG. 1 is a prediction flow chart according to the present invention. As shown in FIG. 1, 4 basic models, lstm, prophet, holt-winter, xgboost, a fusion modelIn which α is i The weights representing each model need to be computed by themselves.
Four models, lstm, xgboost, hold-winter, prophet, can be used for predicting the time series, some of the four models can consider trend, holidays, periodicity and the like, and some of the four models can directly predict future values according to the conventional characteristic machine learning method. Due to the singleness of a single model, the actual effect is not good, and in consideration of the advantages of a combination algorithm, each basic model needs to be fused together, so that the effect of the comprehensive model is guaranteed to be more robust and better.
The single model such as a long-short term memory algorithm (lstm) needs to decompose time-allowed data and is spliced into an input format required by the lstm, the model is fed, xgboost is used for decomposing time and extracting corresponding features, prophet is suitable for single-dimensional time series modeling, factors such as holidays and the like can be considered, the factor is the only one to show holidays, and the last holt-winter is a cubic exponential smoothing algorithm and can consider time series features such as trend and periodicity.
Firstly, four basic models are respectively established, the accuracy of each model is made to be the highest, then a weighted summation mode is used for weighting each model, and then the combined model is used for predicting future data.
The scheme is complex and does not have theoretical support, but the accuracy of the fused model is higher than that of a single model, and the method is similar to a random deep forest or Boosting algorithm, and a plurality of weak learners are combined into a strong learner.
Predicting the telephone traffic in a period of time before based on the data of the first key total telephone traffic of the customer service in the past year, and extracting the telephone traffic in one month to visualize as follows:
fig. 2 shows traffic data for january of a year. The above is an example of the input traffic of customer service in january, one sample is taken every 15 minutes, when prediction is made, the data of historical traffic of two years is used for predicting the traffic of the customer service in the next month, and the purpose of predicting the traffic is to reasonably arrange the working time of each customer service staff for the purpose of scheduling seats.
And (3) stability analysis:
see figure 3 for stability analysis. Wherein the blue line represents the original traffic timing graph from 1 month in 18 years to 2019 year 1, the red line represents the moving average line, and the black line represents the moving average standard deviation.
The traffic data in 2018 are found to be very stable through DF-TEST TEST, and the data in all years are basically stable, so the data set is very suitable for time series modeling for prediction analysis.
The goal is to predict the data for the next 45 days from the data before month 15 and extract the predicted data for the next month, the requirement for the model is that the predicted values must be in the interval of 98.4% and 108.4% of the actual values.
The predicted results are as follows:
FIG. 4holt-winter time series prediction, FIG. 5 lstm model time series prediction, FIG. 6 xgboost model time series prediction, FIG. 7 prophet model time series prediction, above results for 4 models.
The following figures are the results of the final fusion model: fig. 8 shows the final prediction of the 4 models fused together. The central point is the correct point for prediction, and the prediction result is much better than that of manual work.
The foregoing shows and describes the general principles, essential features, and advantages of the invention. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, which are given by way of illustration of the principles of the present invention, and that various changes and modifications may be made without departing from the spirit and scope of the invention as defined by the appended claims. The scope of the invention is defined by the appended claims and equivalents thereof.
Claims (6)
1. The method for predicting the customer service telephone traffic is characterized by comprising the following steps
1) Data pre-processing
Filling missing values, and not deleting abnormal values;
2) Data model building
Respectively establishing four basic models of lstm, holt-winter, xgboost and prophet, wherein the data processing of each model is consistent, but the input data formats of the models are different, and the data preprocessing relates to missing value filling and feature construction;
3) Establishing a linear program
The objective is to predict the data of 45 days in the future according to the historical data and extract the predicted data of the next month, the requirement for the model is that the predicted value must be in the interval of 98.4% and 108.4% of the actual value, and the following linear programming problem is established according to the prediction result of the basic model:
knowing that X is the historical data of the input, f i Is each basic model, Y is the actual traffic data, α i The weight of each of the basic models is represented,
minb＝|∑α i f i (X)-Y|
0.984Y≤b≤1.804Y
the linear programming is established to obtain alpha i Then by sigma alpha i f i (X) as a final model to predict future data.
2. The method of predicting customer service traffic volume of claim 1, wherein: in step 2), the lstm model and the xgboost model adopt rolling prediction and are trained on gpu.
3. The method of predicting customer service traffic volume of claim 1, wherein: in the step 2), the Prophet model adds holidays as factor parameters.
4. The method of predicting customer service traffic volume of claim 1, wherein: in the step 2), adding a seasonal variable to the holt-winter model on the basis of the second exponential smoothing.
5. The method of predicting customer service traffic volume of claim 1, wherein: in step 2), the time series prediction often has the problem of lag term, the reason for generating the lag term is that the sequence has autocorrelation, and differential operation is needed in the processing process, that is, the difference value between the current time and the previous time is used as a regression target.
6. The method of predicting customer service traffic volume of claim 1, wherein: in the step 2), a transverse prediction mode and a longitudinal prediction mode are adopted to predict the same time of a certain day in the future, and then the transverse prediction result and the longitudinal prediction result are averaged; the model training is carried out according to the normal telephone traffic sampled every quarter clock in the horizontal direction, and the model training is carried out according to the telephone traffic sampled every day at the same time in the vertical direction.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211051280.XA CN115456260A (en) | 2022-08-31 | 2022-08-31 | Customer service telephone traffic prediction method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211051280.XA CN115456260A (en) | 2022-08-31 | 2022-08-31 | Customer service telephone traffic prediction method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115456260A true CN115456260A (en) | 2022-12-09 |
Family
ID=84301159
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211051280.XA Pending CN115456260A (en) | 2022-08-31 | 2022-08-31 | Customer service telephone traffic prediction method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115456260A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116233312A (en) * | 2023-05-06 | 2023-06-06 | 广东电网有限责任公司 | Regression-decomposition-based power grid customer service traffic prediction method |
-
2022
- 2022-08-31 CN CN202211051280.XA patent/CN115456260A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116233312A (en) * | 2023-05-06 | 2023-06-06 | 广东电网有限责任公司 | Regression-decomposition-based power grid customer service traffic prediction method |
| CN116233312B (en) * | 2023-05-06 | 2023-08-08 | 广东电网有限责任公司 | Regression-decomposition-based power grid customer service traffic prediction method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109615226B (en) | Operation index abnormity monitoring method | |
| CN111210093B (en) | Daily water consumption prediction method based on big data | |
| CN112910690A (en) | Network traffic prediction method, device and equipment based on neural network model | |
| CN111178585A (en) | Fault reporting amount prediction method based on multi-algorithm model fusion | |
| CN114048436A (en) | Construction method and construction device for forecasting enterprise financial data model | |
| CN111738504A (en) | Enterprise financial index fund amount prediction method and device, equipment and storage medium | |
| CN112990587A (en) | Method, system, equipment and medium for accurately predicting power consumption of transformer area | |
| CN115456260A (en) | Customer service telephone traffic prediction method | |
| CN114897224A (en) | Urban carbon emission prediction method based on mathematical model | |
| CN117236485B (en) | Method for reducing prediction power supply quantity error of power grid based on CNN-BILSTM algorithm | |
| CN116629686A (en) | Method and device for evaluating enterprise energy consumption data | |
| CN116611785A (en) | Power transmission and transformation project cost model construction method, system, equipment and medium based on big data | |
| CN111369048A (en) | Telephone traffic prediction method | |
| CN116562454A (en) | Manufacturing cost prediction method applied to BIM long-short-time attention mechanism network | |
| CN115689331A (en) | Power transmission and transformation project quantity rationality analysis method based on MLP | |
| CN110648023A (en) | Method for establishing data prediction model based on quadratic exponential smoothing improved GM (1,1) | |
| CN115470970A (en) | Electric quantity prediction method and system based on prophet-LightGBM combined model | |
| CN114154696A (en) | Method, system, computer device and storage medium for predicting fund flow | |
| CN112348275A (en) | Regional ecological environment change prediction method based on online incremental learning | |
| CN114065460A (en) | Model processing method, storage medium and electronic device in thermal power generation system | |
| CN111738506A (en) | Cash center cash stock usage amount prediction method and device, electronic device, and medium | |
| JPH11126102A (en) | Predictive method for demand of gas | |
| CN116227714B (en) | Travel mode selection prediction and analysis method and system | |
| CN113723775B (en) | Enterprise and industry operation risk assessment method based on power big data | |
| CN117009752A (en) | Power consumption demand prediction method based on POA-VMD-SSA-KELM model |
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 |