TWI651666B - Personnel configuration method - Google Patents

Abstract

A method for staffing, which is applied to assign one of a plurality of working positions to be assigned, and is executed by an electronic device including a storage module, a processor, and an output module that are signally connected to each other. The method includes: using processing The processor obtains a plurality of historical data from the storage module, each historical data includes at least one characteristic information of the assigned staff; using the processor to establish a survival probability model based on the historical data; according to the at least one characteristic information of the staff to be assigned, using the processor to The survival probability model obtains the survival probability of the person to be assigned in the aforementioned work position; according to the survival probability of the person to be assigned in the aforementioned work position, the processor is used to determine the work position to be assigned to the person to be assigned; and the output module is used to output the position of the person to be assigned. Configured working position.

Description

Staffing approach

The present invention relates to a method for staffing, and in particular, to a method for predicting the survival probability of a person assigned to a work position in a factory.

In production management, the pros and cons of personnel transfer allocation will affect the survival probability of employees (that is, on-the-job probability), and then affect the production capacity of the factory. However, there are no rules for the way people are moved today; in other words, when there are vacancies in the workplace of the factory, they are randomly selected from the new hires or based on subjective ideas (for example: this workplace is subjectively more suitable for men). An employee goes to work at this location. Therefore, it is difficult to predict the survival probability of this employee at the working position, which makes the factory's production capacity unstable.

In view of this, there is a need for a method of staffing and, based on its prediction results, to arrange appropriate new entrants to the vacant workplaces of the plant.

An embodiment of the present invention provides a method for staffing, which is applied to assigning a person to be assigned to one of a plurality of work positions. The method for staffing is performed by an electronic device. The electronic device includes a storage module, A processor and an output module, a storage module, a processor and an output module are signally connected to each other. The method includes: obtaining a plurality of historical data corresponding to each work position, each historical data including at least one characteristic of an assigned worker. Information; using the processor to establish a survival probability model based on historical data; based on at least one characteristic information of the person to be assigned, using the processor to obtain a survival probability of the person to be assigned in each of the work positions using the survival probability model; The survival probability of the assigned personnel in the aforementioned working position, the processor is used to determine the working position to be allocated to the person to be allocated; and the output module is used to output the working position to be allocated to the person to be allocated.

In one embodiment, the use of the processor to determine the work positions to be assigned by the processor further includes: using a linear programming result based on the survival probability, the use of the processor to determine the work positions to be assigned by the processor.

In one embodiment, using the processor to establish a survival probability model based on historical data includes: using the processor to separately extract a plurality of sample groups from the historical data according to different conditions of at least one characteristic information; using the processor to divide each sample group into Plural training data and plural verification data; according to the training data in each sample group, a processor is used to establish a candidate model; according to at least one of the verification data in each sample group, the processor is used to calculate one of the corresponding candidate models. Accuracy; and using the processor to select the candidate model with the highest accuracy as the survival probability model. In one embodiment, the accuracy is obtained by a c-index algorithm. In an embodiment, the condition is the range of historical data, the gender information of the assigned person, the age information of the assigned person, the accommodation information of the assigned person, or a combination of any two or more of the above.

In one embodiment, using the processor to establish a survival probability model based on historical data includes: using the processor to extract complex training data from the historical data; using the processor to execute a decision tree algorithm to establish a survival probability model based on the training data, where the decision is made The tree algorithm includes a plurality of decision trees. Each decision tree includes at least one branch. Each branch includes a node and two leaf nodes extended by a binary classification of at least one feature information. The feature information of the binary classification is different from each other.

In one embodiment, using a processor to execute a decision tree algorithm to build a survival probability model with training data includes: performing a verification algorithm at the same root node with the processor to calculate a verification value for each feature information using the training data. Using the processor to select the feature information corresponding to the largest of the test values as a selected feature information; and using the processor to perform binary classification on the root node based on the selected feature information. In addition, in one embodiment, the verification algorithm is a logarithmic-level verification algorithm.

In an embodiment, using the processor to establish a survival probability model based on historical data includes: using the processor to extract a plurality of sample groups from the historical data respectively according to different conditions of at least one characteristic information; using the processor to divide each sample group into Training data and complex verification data; using the processor to execute a decision tree algorithm to build a candidate model based on the training data of each sample group, the decision tree algorithm includes a plurality of decision trees, each decision tree includes at least one branch, each The branch includes a node and a two-leaf node extended by performing a binary classification on at least one of the characteristic information, and the characteristic information of the binary classification performed by each root node is different; according to at least one of the verification data in each sample Or, the processor calculates an accuracy of the corresponding candidate model; and the processor selects the candidate model with the highest accuracy as the survival probability model. In an embodiment, the condition is the range of historical data, the gender information of the assigned person, the age information of the assigned person, the accommodation information of the assigned person, or a combination of any two or more of the above.

In one embodiment, using the processor historical data to establish a survival probability model includes: using the processor to divide the historical data into complex training data and complex verification data; using the processor to execute a decision tree algorithm to match the training data with different A candidate model is established for each parameter of the decision tree model. The decision tree algorithm includes a plurality of decision trees. Each decision tree includes at least one branch. Each branch includes a node and a binary classification of one of the at least one feature information. The extended two-leaf nodes have different feature information for binary classification performed by each root node; based on at least one of the verification data, the processor calculates an accuracy of the candidate model separately; and uses the processor to select the candidate with the highest accuracy The model serves as a survival probability model. The model parameters are the number of binary classifications, the number of decision trees, the minimum amount of data of the last leaf node in each decision tree, the number of layers in each decision tree, or a combination of any two or more of the above.

In summary, through the staffing method provided in the embodiment of the present invention, it is possible to systematically configure new personnel to an appropriate working position, thereby improving the working probability of the personnel in the working position.

Please refer to FIG. 1, which is a schematic flowchart of a staffing method according to an embodiment of the present invention, which specifically discloses the staffing method of the present invention. The method can be applied to predict the survival probability of a person to be assigned in one of a plurality of work positions. Here, the working position refers to a station in a factory, but it is not limited thereto, and it may also be a station in one of a plurality of workstations in the factory.

As shown in FIG. 8, the staffing method is executed by an electronic device 80. The electronic device 80 includes a storage module 81, a processor 82, and an output module 83. The storage module 81, the processor 82, and the output module 83 are connected to each other through a wired or wireless signal. For example, the electronic device 80 may be an industrial computer, a personal computer, a notebook computer, a smart phone, a tablet computer, or the like. Here, the storage module 81 may be implemented by one or more storage elements. Each storage element may be, for example, a non-volatile memory, a hard disk, an optical disk, or a magnetic tape, but it is not limited thereto. The processor 82 may be implemented by one or more processing elements. Here, each processing element may be a microprocessor, a microcontroller, a digital signal processor, a microcomputer, a central processing unit, a field programmable gate array, a programmable logic device, a state device, a logic circuit, an analog circuit, a digital circuit, and / Or any device that operates signals (analogs and / or digits) based on operation instructions, but it is not limited here. The output module can be a screen, printer, or voice output device (such as a speaker), but it is not limited here.

In step S10, the processor 82 is used to obtain the plurality of historical data of each work position from the storage module 81. Table 1 shows the historical data of a work place in the past two years. It includes the characteristic information (gender, age, whether or not accommodation) of 10 workers who have worked in this work place and their individual survival time information (on-the-job Days, whether you have left). Among them, the "Gender" field, "Whether or not to stay" field, and "Whether to leave" field are distinguished by 1 and 0 (male / female, accommodation / non-accommodation, employment / resignation). The number of working days refers to the number of working days of the staff member in this working position, and whether the employee has left the service means whether the staff member has left the working position and no longer works. In other words, the time-to-live information includes a duration during which a worker is set in the work location. It should be noted that the information provided in Table 1 is only an example, and the present invention is not limited thereto. For example, the characteristic information of the staff is not limited to the age, gender, whether or not they are staying; the historical data is not limited to two years, and can also include the staff of this work location and the staff of the nearby work location. Information such as gender ratio, staff education, or marital status of staff. In addition, the work content of the plural work positions classified as the same station is not required to be completely the same. For example, the work position can be based on the working time of the work position, the operation operation of the person, and the operation posture of the person (sitting / standing position). The similarity of position-related parameters considers several working positions with higher similarity as the same station.

Table 1

In step S20, the processor 82 is used to establish a survival probability model based on the historical data. The type of the survival probability model is not limited here, as long as the survival probability can be accurately predicted. However, the accuracy of the survival probability model will be affected by the type of model and the data parameters used in building the model. In an embodiment, in order to make the finally adopted survival probability model have higher accuracy to generate a reliable prediction result, in step S20, the accuracy of the survival probability model may be further calculated. As shown in FIG. 2, the data parameters used and the parameters of the survival probability model can be tested by calculating the accuracy by the following steps to ensure that the final survival probability model can accurately predict the survival probability (ie, has the highest accuracy).

First, in step S201, the processor 82 is used to extract a plurality of sample groups from the historical data according to different conditions of at least one feature information. In an embodiment, the condition described in step S201 may be the age range of the staff; it is assumed that this work position requires heavy labor and only considers historical data of staff under 30, 28, and 25 Therefore, three different sample groups corresponding to the above three conditions can be obtained from Table 1. In another embodiment, the condition mentioned in step S201 may be the time range of historical data; in detail, the historical data within one year and half a year may be obtained from Table 1 corresponding to the above two periods, respectively. Different sample groups. In addition, the condition can also be a range combination of several characteristic information; for example, suppose that this work position is heavy labor and requires normal overtime, and only considers those who have accommodation and are under 30, under 28, or under 25 For the historical data of the staff, three different sample groups corresponding to the above three ages can be obtained from Table 1. In addition, the condition can also be the selection of the range after the value of the characteristic information is processed through variables; for example, the age field in the historical data is first processed by rounding, and then a sample group older than 25 years old is selected. Sample group older than 28 years.

In step S202, the processor 82 is used to divide each sample group into training data and verification data. The training data is used for the subsequent steps to establish the survival probability model, and the verification data is used to verify the accuracy of the survival probability model. Here, the sampling method is not limited; for example, it can be a bootstrap method that needs to be replaced after extraction or a sampling method that cannot be replaced after extraction. In addition, the number of training data and verification data (such as the number of personnel) can be the same, that is, the sample group is divided into training data and verification data; or the number of training data and verification data can be different. It should be noted that the training data and verification data can be directly determined through the self-service sampling method; in other words, in the self-service sampling method, the data that has not been extracted is the verification data.

In step S203, a candidate model is separately established by using the processor 82 according to the training data of each sample group. Taking the foregoing three ages as an example, three candidate models are respectively established based on the training data of the sample group of different age ranges. .

In step S204, the processor 82 is used to calculate an accuracy of the corresponding candidate model according to at least one of the verification data in each sample group. In this example, the aforementioned three candidate models are respectively used to calculate the accuracy of the model through their respective verification data. Because these verification data are also from the same historical data as the verification data for the candidate model, the estimated survival probability generated by incorporating the feature information in the verification data into the candidate model can be compared with the actual survival time information in the verification data. By comparison, the accuracy of the survival probability model can be quantified. In an embodiment, the accuracy of the survival probability model can be estimated through the c-index algorithm (the closer the value of c-index is to 1, the more accurate).

After that, in step S205, the processor 82 may be used to select the candidate model with the highest accuracy as the survival probability model, so as to facilitate subsequent assessment of the survival probability of the new entrant.

Then, in step S30, based on the characteristic information of a person to be assigned, the processor 82 can be used to obtain the survival probability of each person in these working positions by using the above-mentioned survival probability model. Next, in step S40, the processor 82 is used to determine the work position to be assigned by the processor according to the survival probability of the person to be assigned at the above-mentioned work position. Finally, in step S50, the output module 83 is used to output the working position to be allocated by the person to be assigned.

Please refer to FIG. 3, which is another detailed flowchart of step S20 according to another embodiment of the present invention. The difference from the embodiment listed in FIG. 2 is that step S203A in FIG. 3 is a specific embodiment of step S203. That is, the candidate tree is established by using the processor 82 to execute a decision tree algorithm. For example, the decision tree algorithm can be a random survival forest (RSF) algorithm, but it is not limited to this.

In the decision tree algorithm, a candidate model is built using a complex decision tree. In detail, a plurality of samples are taken from each sample group to obtain training data of a complex array, and then a decision tree is generated based on each group of training data. Therefore, for each sample group, a plurality of decision trees can be generated. It should be noted that when the random survival forest (RSF) algorithm is used to build the candidate model, because the random survival forest algorithm uses the self-help sampling method to obtain the training data, the processor 82 may Repeat the sampling of the same data in the sample group (repeated sampling), so the unsampled data in each sample group will become the verification data, so the training data and verification data can be generated through the self-service sampling method. For example, for the same group of 10 data, the processor 82 samples the 10 data in a self-help sampling method in each sample, sampling 5 times in total, so a total of 5 decision trees are generated. Because there is usually repeated sampling when sampling by the self-help sampling method, the number of data items for each sampling is not necessarily the same, so the sampling data (that is, training data) used by each decision tree is not necessarily the same. Each decision tree includes at least one branch. In addition, the feature information of the binary classification performed by each root node in the same decision tree is different from each other. As shown in FIG. 4, in an embodiment, the decision tree 50 includes three branches 51/52/53, and the branch 51 includes a node 51a and a two-leaf node 51b extended by performing a binary classification on at least one feature information. 51c. Similarly, branch 52 includes a node 52a (ie, leaf node 51b) and two leaf nodes 52b, 52c extended by performing a binary classification on at least one feature information; branch 53 (ie, leaf node 51c) includes a node 53a and The two-leaf nodes 53b and 53c extended by performing a binary classification on at least one feature information. The feature information of the binary classification performed by each root node 51a / 52a / 53a in the same decision tree is different from each other. For example, in a sample, 6 out of a total of 10 data were extracted; in Figure 4, the root node 51a represents the 6 training data, and the path between the root node 51a and the leaf nodes 51b, 51c is Represents the binary classification of the 6 training data based on the feature information of whether to stay. In other words, during the establishment of this branch, the 6 training data are classified according to the characteristic information of whether or not they stay. If the training data is that there is a resident, it is classified as the leaf node 51b. Classified to leaf node 51c. As shown in Figure 4, after classification, the number of data with accommodation is 2, and the number of data without accommodation is 4. Here, the department continues to perform binary classification on the two sets of data. In other words, in this embodiment, the leaf node 51b is the root node 52a of the branch 52, and the path between the root node 52a and the leaf nodes 52b and 52c indicates whether the two training materials for the root node 52a are based on whether they are age Binary classification greater than or equal to 27.6 years old; similarly, leaf node 51c is the root node of branch 53, and the path between root node 53a and leaf nodes 53b and 53c represents 4 training data for root node 53a Binary classification based on gender-specific information. After the aforementioned binary classification, as shown in the figure, the number of data of leaf node 52b is 1 (age is greater than or equal to 27.6 years old), while the number of data of leaf node 52c is 1; the number of data of leaf node 53b is 2 (gender is Male), and the number of leaf nodes 53c is 2. It should be noted that although a single decision tree has three root nodes as an example, the present invention is not limited thereto.

In addition, although FIG. 4 is based on the order of whether to stay, sex, and age is 27.6 years or older as a classification basis, the present invention is not limited thereto. Furthermore, in an embodiment, the basis and order of classification can be determined by a verification algorithm. The flow shown in FIG. 5 is used to describe the foregoing step S203A, which includes: step S2031A, step S2032A, and step S2033A. Step S2031A: At the same root node, use processor 82 to execute a verification algorithm to calculate a verification value for each feature information using training data; step S2032A: use processor 82 to select the feature information corresponding to the largest of the verification values as a Selected feature information; and step S2033A: use the processor to perform binary classification on the root node based on the selected feature information. Classification of the same set of training data with different types of feature information will produce different test values. For example, the test value obtained by classifying the foregoing 6 training materials based on whether to stay is 5 and the test value obtained by classifying the foregoing 6 training materials based on gender is 2. In another embodiment, the same set of training data can be classified with different values of feature information of the same category, and different test values will also be generated. For example, whether the age is greater than or equal to 27.6 years old for the aforementioned 6 training data The test value obtained by classifying is 8 and the test value obtained by classifying the aforementioned 6 training materials based on whether they are 22.7 years or older is 0.7. The test value is determined by the difference between the data of the two leaf nodes after classification. When the difference between the data of the two leaf nodes is higher, it means that classification according to this classification rule is appropriate, so it has a higher test. value. In one embodiment, the verification algorithm is a log-rank score algorithm.

According to the above steps, after a decision tree is generated, the system can use the Nelson-Aalen method to calculate a survival function for each leaf node at the extreme end of the decision tree according to the training data in it. The survival function is a function of time, which shows that the survival probability decreases with time; that is, the survival function is only related to the survival time information of the training data. Therefore, through the survival function, it is possible to estimate the survival probability of a new recruit on the Nth day after taking office. The specific method is as follows. After the decision tree is generated, the characteristic information of the newcomer is substituted into this decision tree; in other words, the characteristic information of the newcomer is classified according to the classification rules of the decision tree, and it can be known to which end leaf the person will be classified. node. Then, the survival function corresponding to the end-most leaf node is used to estimate the survival probability on the Nth day after employment. Because there are multiple decision trees in this decision tree algorithm, and because each decision tree is generated based on different training data, the classification rules used in each decision tree are not necessarily the same as their order. In this way, a plurality of different survival probability estimates can be obtained, and finally these estimates are averaged to obtain the average survival probability of the new recruit at the work position on the Nth day after employment.

Because the decision tree itself contains the parameters of the decision tree model, such as the number of decision trees, the minimum amount of data on the leaf nodes at the end of the decision tree, and the number of layers in the decision tree (in the case of Figure 4, it is 2 layers). The survival probability model has high accuracy. In another embodiment, these decision tree model parameters are selected through accuracy (such as the c-index algorithm described above). Please refer to FIG. 6. Compared to FIG. 3, the method of this embodiment does not extract sample groups from historical data according to the feature information, but can directly use historical data as the sample group, omitting the foregoing step S201, and using a different method. The decision tree model parameters are combined with the training data to establish candidate models (step S203B). That is, each candidate model uses the same training data, but uses different decision tree model parameters. The remaining steps S202B, S204B, and S205 are substantially the same as those in FIG. 3 described above, and will not be repeated here.

In an embodiment, the processor 82 may also be used to extract a plurality of sample groups from historical data according to different conditions of at least one feature information, and then use the processor 82 to execute a decision tree algorithm to use the training data of each sample group. A candidate model is established with different decision tree model parameters. Then, according to the verification data of each sample group, the accuracy of each candidate model is calculated by the processor 82 to determine the survival probability model.

Table 2

table 3

Table 4

The situation described in the above paragraph is to predict the survival probability of a person to be assigned at the Nth day of the work position. On the other hand, if the number of persons to be assigned and the number of work positions need to be predicted to increase, the above-mentioned method can be used to obtain the survival probability of each person to be assigned in each work position on the Nth day (see Table 2). After that, the processor 82 determines the work position to be assigned by the person to be assigned. For example, the optimal configuration between the person to be assigned and the work position can be obtained by performing a linear programming with the processor 82 (see Table 4). In other words, in an embodiment, as shown in FIG. 7, after step S30, the method further includes: step S41: using the processor 82 to determine the working position to be assigned to the person to be allocated according to a linear programming result according to the survival probability; Table 2 shows the prediction of the survival probability of 5 employees and 3 positions on the 50th day. Table 3 is the result of algebraizing Table 2, and Table 4 is a linear programming calculation on Table 2. After that, the best solution is obtained. 1 in Table 4 indicates that the person to be assigned is finally located in this working position, and 0 is not. The linear programming algorithm can be a branch and bound (Bnb) algorithm, where one or more constraints can be added. For example: the sum of all x values in each row is less than or equal to 1 (that is, only one person in a work position); the sum of all x values in each column is less than or equal to 1 (that is, one person can only be arranged To a working position). With this, as in Table 4, an optimal configuration can be obtained to facilitate the staffing of production management.

In summary, through the staffing method provided by the embodiment of the present invention, it is possible to systematically configure new personnel to an appropriate working position, thereby improving the working probability of the personnel in the working position.

Although the above embodiments disclose specific embodiments of the present invention, they are not intended to limit the present invention. Those with ordinary knowledge in the technical field to which the present invention pertains should not deviate from the principles and spirit of the present invention. Various changes and modifications can be made to it, so the scope of protection of the present invention shall be defined by the scope of the accompanying patent application.

50‧‧‧ Decision Tree

51‧‧‧ branch

52‧‧‧ branch

53‧‧‧ branch

51a‧‧‧root node

52a‧‧‧root node

53a‧‧‧root node

51b‧‧‧ leaf node

51c‧‧‧ leaf node

52b‧‧‧ leaf node

52c‧‧‧ Leaf Node

53b‧‧‧ leaf node

53c‧‧‧ Leaf Node

S10‧‧‧ Use the processor to obtain multiple historical data corresponding to each work position from the storage module

S20‧‧‧ uses the processor to build a survival probability model based on historical data

S201‧‧‧ uses the processor to extract a plurality of sample groups from the historical data according to different conditions of at least one characteristic information

S202‧‧‧ use the processor to divide each sample group into complex training data and complex verification data

S202B‧‧‧ uses the processor to divide the historical data into plural training data and plural verification data

S203‧‧‧ uses the processor to establish a candidate model according to the training data in each sample group

S203A‧‧‧ uses the processor to execute a decision tree algorithm to build a candidate model based on the training data of each sample group

S203B‧‧‧ uses a processor to execute a decision tree algorithm, and uses training data to match different decision tree model parameters to establish a candidate model

S2031A‧‧‧ uses the processor to execute a verification algorithm on the same node to calculate a verification value for each feature information using the training data

S2032A‧‧‧Selects the feature information corresponding to the largest one of the test values as a selected feature information

S2033A‧‧‧ uses the aforementioned selected feature information to use the processor to perform binary classification on the root node

S204‧‧‧ uses the processor to calculate an accuracy of the corresponding candidate model according to at least one of the verification data in each sample group

S204B‧‧‧ uses the processor to calculate an accuracy of the candidate model according to at least one of the verification data

S205‧‧‧ uses the processor to select the candidate model with the highest accuracy as the survival probability model

S30‧‧‧ According to the characteristic information of a person to be assigned, using a processor to obtain a survival probability of the person to be assigned in a plurality of working positions by using a survival probability model

S40‧‧‧ According to the survival probability of the personnel to be assigned in the above working position, the processor is used to determine the working position to be assigned to the person to be assigned

S41‧‧‧ Use the result of a linear programming based on the survival probability of the processor to determine the work position to be assigned to the assigned staff

S50‧‧‧ Use the output module to output the working position to be assigned

80‧‧‧ electronic device

81‧‧‧Storage Module

82‧‧‧ processor

83‧‧‧Output Module

[FIG. 1] It is a schematic flowchart of a personnel deployment method according to an embodiment of the present invention. [FIG. 2] A detailed flowchart of step S20 of the method for staffing according to an embodiment of the present invention. [FIG. 3] It is another detailed flowchart of step S20 of the method for staffing according to an embodiment of the present invention. 4 is a schematic diagram of a decision tree of a method for staffing according to an embodiment of the present invention. [FIG. 5] A detailed flowchart of step S203A of the method for staffing according to an embodiment of the present invention. [FIG. 6] It is another detailed flowchart of step S20 of the method for staffing according to an embodiment of the present invention. 7 is a schematic flowchart of a personnel deployment method according to another embodiment of the present invention. [FIG. 8] A functional block diagram of an electronic device applied to perform a method for staffing according to an embodiment of the present invention.

Claims (11)

A staffing method is applied to assign a person to be assigned to one of a plurality of work positions. The staffing method is executed by an electronic device, the electronic device includes a storage module, a processor, and a An output module, the storage module, the processor, and the output module are signally connected to each other. The method includes: using the processor to obtain a plurality of historical data corresponding to each of the working positions from the storage module, and each of the historical data includes At least one feature information of an assigned worker; using the processor to establish a survival probability model from the historical data; and using the processor to obtain the waiting probability using the survival probability model based on the at least one feature information of the to-be-assigned staff A survival probability of the assigned personnel at the working positions respectively; using the processor to determine the working position to which the person to be assigned should be assigned according to the survival probability of the personnel to be assigned at the working positions; and using the output mode The group outputs the working position to which the person to be assigned should be assigned. The method for staffing as described in claim 1, wherein using the processor to determine the working position to which the person to be assigned further includes: using the processor to determine the result of a linear programming based on the survival probability The working position to be assigned by the person to be assigned. The staffing method according to claim 1, wherein the using the processor to establish a survival probability model from the historical data includes: using the processor to retrieve the historical data from the historical data according to different conditions of the at least one characteristic information. A plurality of sample groups are extracted respectively in the sample; each of the sample groups is divided into a plurality of training data and a plurality of verification data by the processor; according to the training data in each of the sample groups, a candidate model is separately established by the processor; according to each At least one of the verification data in the sample group, using the processor to calculate an accuracy of the corresponding candidate model; and using the processor to select the candidate model with the highest accuracy as the survival probability model. The method for staffing according to claim 1, wherein using the processor to establish a survival probability model from the historical data includes: using the processor to retrieve plural training data from the historical data; and using the processor A decision tree algorithm is executed to establish the survival probability model with the training data, wherein the decision tree algorithm includes a plurality of decision trees, each decision tree includes at least one branch, each branch includes a node, and the at least one The feature information is a two-leaf node extended by a binary classification, and the feature information of the binary classification performed by each root node in the same decision tree is different from each other. The method for staffing as described in claim 4, wherein using the processor to execute a decision tree algorithm to establish the survival probability model with the training data includes: using the processor to perform a test at the same root node The algorithm uses the training data to calculate a test value for each of the feature information; uses the processor to select the feature information corresponding to the largest of the test values as a selected feature information; and uses the selected feature information Use the processor to perform the binary classification at the root node. The staffing method according to claim 5, wherein the verification algorithm is a logarithmic-level verification algorithm. The staffing method according to claim 1, wherein the using the processor to establish a survival probability model from the historical data includes using the processor to extract the historical data from the historical data according to different conditions of the at least one characteristic information. Extract complex sample groups separately; use the processor to divide each sample group into complex training data and complex verification data; use the processor to execute a decision tree algorithm to create a candidate based on the training data of each sample group Model, the decision tree algorithm includes a plurality of decision trees, each of the decision trees includes at least one branch, each branch includes a node, and two leaves extended by performing a binary classification on one of the at least one feature information Nodes, the feature information of the binary classification performed by each root node is different; according to at least one of the verification data in each of the samples, using the processor to calculate an accuracy of the corresponding candidate model; And using the processor to select the candidate model with the highest accuracy as the survival probability model. The staffing method as described in claim 3 or 7, wherein the condition is the range of the historical data, the gender information of the assigned person, the age information of the assigned person, the accommodation information of the assigned person, or any of the above. A combination of more than two. The method for staffing according to claim 1, wherein using the processor to establish a survival probability model with the historical data includes: using the processor to divide the historical data into plural training data and plural verification data; using The processor executes a decision tree algorithm, and uses the training data and different decision tree model parameters to respectively establish a candidate model. The decision tree algorithm includes a plurality of decision trees, each of which includes at least one branch, each The branch includes a node and a two-leaf node extended by performing a binary classification on one of the at least one feature information, and the feature information of the binary classification performed by each root node is different; according to the verification data Using at least one of the processors to separately calculate an accuracy of the candidate models; and using the processor to select the candidate model having the highest accuracy as the survival probability model. The method for staffing as described in claim 9, wherein the parameters of the decision tree model are the number of times of the binary classification, the number of the decision trees, the minimum amount of data of each leaf node in the decision tree, and each of the The number of levels of the decision tree or a combination of any two or more of the above. The staffing method as described in claim 3, claim 7, or claim 9, wherein the accuracy is obtained by a c-index algorithm.