TWI807780B - Turnover rate prediction method and electronic apparatus thereof - Google Patents

Abstract

A turnover rate prediction method and an electronic apparatus thereof are provided. The turnover rate prediction method includes the following steps. An employee data set corresponding to employees is obtained. The employees are grouped in multiple subsets based on the employee data set and an influencing factor, and multiple survival curves corresponding to the subsets are calculated respectively. The survival curves are grouped in multiple curve groups based on similarity of the survival curves to obtain a plurality of curve groups. A plurality of clustered survival curves corresponding to the curve groups are calculated respectively, and at least one segmentation time point is obtained based on slope changes of the clustered survival curves. A proportional hazard model is used to calculate a turnover rate prediction result of the employees corresponding to the influencing factor based on the segmentation time point.

Description

Turnover rate prediction method and its electronic device

The disclosure relates to a statistical analysis mechanism, and in particular to a turnover rate prediction method and its electronic device.

In general, turnover rates vary with seniority or other important variables. Therefore, it is not possible to obtain objective evaluation results if the turnover rate is judged directly based on a variable factor. For example, the turnover rate of housing employees is many times higher than that of non-housing employees. But in fact, the turnover rate will change with the length of employment. For example, it is easier to leave a job without a dormitory at the beginning, but after a certain period of time, having a dormitory will have little effect on whether to leave the job. Accordingly, continuous variables need to be categorized, and how to properly divide them is a factor to be considered.

This disclosure provides a turnover rate prediction method and its electronic device, which can improve the reference value of the analysis results.

The turnover rate prediction method disclosed in this disclosure is suitable for execution by a processor, the said The turnover rate prediction method includes: obtaining employee data sets corresponding to a plurality of employees; grouping the employees into multiple subsets according to the employee data sets and influencing factors, and calculating multiple survival curves respectively corresponding to these subsets; grouping the survival curves into multiple curve groups according to the similarity of the survival curves; calculating the plurality of grouped survival curves corresponding to these curve groups respectively, and obtaining at least a split time point according to the slope change of the survival curve after the grouping; The employee corresponds to the predicted result of the turnover rate of the influencing variable.

The electronic device for survival rate analysis of the present invention includes: a memory storing at least one program code instruction; a processor coupled to the memory to execute at least one program code instruction to implement the survival rate analysis method and turnover rate prediction method.

Based on the above, this disclosure can reasonably divide the variables before evaluating them, thereby improving the reference value of the analysis results.

100: Electronic device

110: Processor

120: storage

121: database

S205~S225: Steps of the turnover rate prediction method

301~306: Survival curve

401~403: Survival curve after clustering

T1, T2: Segmentation time points

FIG. 1 is a block diagram of an electronic device according to an embodiment of the disclosure.

FIG. 2 is a flowchart of a method for predicting turnover rate according to an embodiment of the present disclosure.

FIG. 3 is a schematic diagram of survival curves according to an embodiment of the present disclosure.

FIG. 4 is a schematic diagram of survival curves after clustering according to an embodiment of the present disclosure.

FIG. 1 is a block diagram of an electronic device according to an embodiment of the disclosure. Please refer to FIG. 1 , the electronic device 100 includes a processor 110 and a storage 120 . The processor 110 is coupled to the storage 120 . The storage 120 includes a database 121 and at least one code instruction, and the database 121 stores employee data sets corresponding to a plurality of employees.

The processor 110 is, for example, a central processing unit (Central Processing Unit, CPU), a physical processing unit (Physics Processing Unit, PPU), a programmable microprocessor (Microprocessor), an embedded control chip, a digital signal processor (Digital Signal Processor, DSP), an application specific integrated circuit (Application Specific Integrated Circuits, ASIC) or other similar devices.

The storage 120 is, for example, any type of fixed or removable random access memory (Random Access Memory, RAM), read-only memory (Read-Only Memory, ROM), flash memory (Flash memory), hard disk or other similar devices or a combination of these devices. The storage 120 is used for storing the database 121 and one or more code fragments. After the above code fragments are installed, the processor 110 will execute the following turnover rate prediction method.

FIG. 2 is a flowchart of a method for predicting turnover rate according to an embodiment of the present disclosure. Please refer to FIG. 2 , in step S205 , employee data sets corresponding to a plurality of employees are acquired. Next, in step S210, the employees are grouped into a plurality of subsets according to the employee data sets and influencing variables, and a plurality of survival curves corresponding to the subsets are calculated. In some embodiments, the influencing variable is a variable affecting the turnover rate, and the influencing variable is a continuous variable, such as age, seniority, overtime hours or salary, and other variables with continuous characteristics.

Specifically, the processor 110 divides multiple intervals based on the numerical ranges corresponding to the influencing variables, and divides the employee data set into multiple subsets corresponding to the multiple intervals. In an embodiment, a set value v can be used to divide multiple intervals. The intervals are, for example: <N-2×v, N-2×v~N-v, N-v~N, N~N+v, N+v~N+2×v, >N+2×v, wherein N is a reference value, which is a value greater than 2×v. For example, if N=35000 and v=5000, the following intervals are divided: <25000, 25000~30000, 30000~35000, 35000~40000, 40000~45000, >45000. However, this is only for illustration, not limitation.

Next, the processor 110 divides the employee data set into sub-sets corresponding to each section. For example, assuming that the influencing variable is set to "salary" and the unit is "yuan", the salary can be divided into the following six intervals: <25000 (corresponding to sub-set A1), 25000-30000 (corresponding to sub-set A2), 30000-35000 (corresponding to sub-set A3), 35000-40000 (corresponding to sub-set A4), 40000-45000 (corresponding to sub-set Set A5), >45000 (corresponding to sub-set A6), and the employee data set includes the working days, salary, age, overtime hours and other data of each employee. Based on the salary range above, the employee data set is divided into 6 data groups, namely, the data group whose salary is <25,000, the data group whose salary is between 25,000 and 30,000, the data group whose salary is between 30,000 and 35,000, the data group whose salary is between 35,000 and 40,000, the data whose salary is between 40,000 and 45,000, and the data group whose salary is greater than 450,000. The six data groups are respectively sub-sets A1-A6.

After obtaining the subsets, the processor 110 uses a survival analysis algorithm to calculate a plurality of survival curves corresponding to the subsets. Survival analysis algorithms such as Kaplan-Meier method. In detail, the processor 110 uses a survival analysis algorithm to calculate the survival rates at different time points in the subsets A1 - A6 to obtain corresponding survival curves.

Table 1 is taken as an example below for illustration. In Table 1, corresponding to six intervals (<N-2×v, N-2×v~N-v, N-v~N, N~N+v, N+v~N+2×v, >N+2×v) to obtain sub-sets A1~A6, and calculate the corresponding survival rate at time point t1-tn for each sub-set. Taking the subset A1 as an example, in the data group corresponding to the interval <N-2×v, calculate the survival rate S(A1,t1)~S(A1,tn) corresponding to the time point t1~tn, and the rest can be deduced by analogy to obtain multiple survival rates corresponding to the time point t1~tn in each of the subsets A1~A6, and then obtain the corresponding survival curve.

Below is an example to illustrate how to calculate the survival rate.

Please refer to Table 2. Table 2 is, for example, described with the data group of subset A1 and its corresponding interval<N−2×v. Set multiple time points t1~tn in the section corresponding to the subset A1, and query each time point from the corresponding data group one by one to obtain the number of employees I _t-1 at time point t-1 and the number of people leaving the job d t within time point t-1~time point _t .

Next, use the following formula to calculate the turnover rate H(t) from time point t-1 to time point t: H(t)=d _t /I _t-1 .

Then, the following formula is used to calculate the survival rate S(t) corresponding to the time point t: S(t)=S(t-1)×(1-H(t)), where S(0)=1.

Taking the time point t1 as an example, query the data group (subset A1) in the interval <N-2×v, and obtain the number of employees I _t0 (=300) at the time point t0 (=0) and the number of people leaving the job d _t1 (=98) within the time point t0~t1. After that, the turnover rate (t1)=0.33 (98/300) is calculated. Finally, the survival rate S(t1)=1×(1-0.33)=0.673 was calculated. The values in Table 2 are obtained by analogy.

After calculating the survival rate corresponding to each time point, the corresponding survival curve can be drawn with time as the horizontal axis and the survival rate as the vertical axis. FIG. 3 is a schematic diagram of survival curves according to an embodiment of the present disclosure. In FIG. 3 , the survival curves 301 - 306 correspond to the subsets A1 - A6 respectively.

After the survival curves corresponding to each subset are obtained, in step S215, the survival curves are grouped into a plurality of curve groups according to the similarity of the survival curves. Each curve group here includes at least one survival curve. For example, classification algorithms such as hierarchical clustering or K-means method can be used for clustering.

Referring to FIG. 3 , the survival curves 301 to 306 are respectively regarded as a matrix, which is brought into a classification algorithm for grouping. For example, survival curves 301, 302 (corresponding to subsets A1, A2) are grouped into curve group G1, survival curves 303, 304 (corresponding to subsets A3, A4) are grouped into curve group G2, survival curves 305, 306 (corresponding to subsets A5, A6) are grouped into curve group G3. Therefore, based on the intervals corresponding to each subset, the intervals corresponding to the curve groups G1~G3 can be obtained respectively: <N-v, N-v~N+v, >N+v, as shown in Table 3 below.

After that, in step S220, calculate the plurality of post-group survival curves respectively corresponding to the curve groups, and obtain at least one split time point according to the slope change of the post-group survival curves.

The processor 110 divides the employee data set into curve groups G1-G3 corresponding to respective data groups. Afterwards, a survival analysis algorithm is used to calculate multiple survival rates corresponding to multiple time points in each data group, so as to obtain a grouped survival curve corresponding to each curve group. Here, the calculation of the survival rate can refer to the description in Table 2 above.

Taking the curve group G1 as an example, in the data group corresponding to the interval <N-v, calculate the survival rate S(G1,t1)~S(G1,tn) corresponding to the time point t1~tn, and the rest can be deduced by analogy to obtain multiple survival rates of the curve group G1~G3 corresponding to the time point t1~tn, and then obtain the corresponding survival curve after grouping.

For example, FIG. 4 is a schematic diagram of survival curves after grouping according to an embodiment of the present disclosure. In FIG. 4 , survival curves 401 to 403 after grouping correspond to curve groups G1 to G3 respectively.

After the post-group survival curves 401-403 are obtained, the cut-off time points are found in the post-group survival curves 401-403. In some embodiments, the method for finding the segmentation time point may first use the Moving Average (Moving Average) method to smooth the survival curves 401-403 after clustering, and then calculate the slopes of multiple time points in the smoothed survival curves after clustering, so as to find the time point with the largest slope change as the segmentation time point. For example, in the example shown in Figure 4, the survival curve after clustering In step 401 , the splitting time point T1 is found, and in the survival curve 403 after clustering, the splitting time point T2 is found.

Finally, in step S225, based on the split time point, a proportional hazard model is used to calculate the predicted result of the turnover rate of the employee corresponding to the influencing variable. For example, the corresponding time range is set by using the obtained segmentation time point. Taking the division time points T1 and T2 in FIG. 4 as an example, three time ranges can be set, namely, <T1, T1~T2, >T2. The proportional hazards model is, for example, a Cox proportional hazards model.

The Cox proportional hazards model is a semiparametric regression model that can be used to predict the effect of one or more different variables on survival at a certain time. The formula for the Cox proportional hazards model is as follows:

；t代表存活的時間點；h(t|x)代表在第t個時間點的情況下給定x的風險；h₀(t)表示在第t個時間點時的基礎風險，例如為任意一個基線危險函數(baseline hazard function)；x=I(影響變因，年資)，在符合條件下x=1，在不符合條件下x=0。舉例來說，假設影響變因為“薪資”，假設條件為年資<30，即，I(薪資，年資<30)，則x=1，反之x=0。另，影響變因亦可以為年齡、加班時數等。 Among them, β is the regression coefficient;

;t represents the time point of survival; h(t|x) represents the risk of a given x at the tth time point; h ₀ (t) represents the base risk at the tth time point, for example, any baseline hazard function; x=I (influence variable, seniority), x=1 when the conditions are met, and x=0 when the conditions are not met. For example, assuming that the influencing variable is "salary", the assumed condition is seniority<30, that is, I(salary, seniority<30), then x=1, otherwise x=0. In addition, the influencing factors can also be age, overtime hours, etc.

The formula derivation of the Cox proportional hazards model is expanded as follows, wherein ε is an error term:

First use the employee data set to estimate β (including β ₁ ~ β _k ). For example, β can be estimated by using Maximum Likelihood Estimation (MLE). Then β is substituted into the formula to calculate the hazard ratio HR. The larger β means the higher the turnover risk of variable _xi .

In addition, one of the curve groups can be used as the baseline, and the risk ratios corresponding to the other curve groups are compared with the risk ratios corresponding to the baseline. For example, the following formula can be used to calculate the risk level based on the baseline: HR(x=x ₁ )=exp(β ₁ ).

Among them, β1 is obtained through the calculation of maximum likelihood estimation; when HR>1, it means that the turnover risk of the influencing variable _x1 is higher than the baseline, and when HR<1, it means that the turnover risk of the influencing variable _x1 is lower than the baseline.

As shown in Table 4, the curve group G1 is used as the baseline to obtain how much the other curve groups G2 and G3 are higher than the baseline.

To sum up, this disclosure can reasonably divide the influencing variables before evaluating Therefore, the reference value of the analysis results can be improved. Specifically, this disclosure grouped the estimated survival curves, thereby reducing the irrationality of the artificial grouping of continuous variables, and using the segmentation time points to obtain the variables of the time interval (seniority), and then simultaneously evaluating seniority and other influencing variables to formulate follow-up talent retention policies more effectively.

S205~S225: Steps of the turnover rate prediction method

Claims (10)

A method for predicting turnover rate, suitable for execution by a processor, comprising: obtaining an employee data set corresponding to a plurality of employees; grouping these employees into multiple subsets according to the employee data set and an influencing variable, and using a survival analysis algorithm to calculate a plurality of first survival rates corresponding to multiple time points for the subsets, and then obtaining plural survival curves corresponding to the subsets respectively; grouping the survival curves into multiple curve groups according to the similarity of the survival curves; The employee data set is divided into a plurality of data groups corresponding to the curve groups; using the survival analysis algorithm to calculate multiple second survival rates corresponding to the time points in the data groups to obtain a plurality of grouped survival curves respectively corresponding to the curve groups, and obtain at least a split time point according to the slope changes of the grouped survival curves; The turnover rate prediction method as described in claim item 1, wherein the step of grouping the employees into the subsets according to the employee data set and the influencing variable includes: dividing a plurality of intervals based on the numerical range corresponding to the influencing variable; and dividing the employee data set into the subsets corresponding to the intervals. The turnover rate prediction method as described in Claim 1, wherein the step of obtaining at least a cut-off time point according to the slope change of the grouped survival curves includes: using a moving average method to smooth the grouped survival curves; and finding at least one time point with the largest slope change in the smoothed grouped survival curves as the at least cut-off time point. The turnover rate prediction method as described in claim item 1, wherein the step of using the proportional hazards model to calculate the turnover rate prediction results of the employees corresponding to the influencing variable includes: using the proportional hazards model to calculate the risk ratios of the curve groups in multiple time ranges, wherein the time ranges are set based on the at least one split time point; using one of the curve groups as a baseline, comparing the risk ratios corresponding to the other curve groups with the risk ratio corresponding to the baseline; and using the comparison results as the turnover rate predict the outcome. The method for predicting turnover rate as claimed in item 1, wherein the influencing variable is a continuous variable. The method for predicting turnover rate as described in Claim 1, wherein the influencing variable includes one of age, seniority, overtime hours and salary. An electronic device for predicting turnover rate, comprising: a memory storing an employee data set corresponding to a plurality of employees and at least one code instruction; a processor coupled to the memory to execute the at least one code instruction to implement a turnover rate prediction method, the processor is used for: Obtain the employee data set; group these employees into multiple subsets according to the employee data set and an influencing factor, and use a survival analysis algorithm to calculate a plurality of survival curves corresponding to the subsets; group the survival curves into multiple curve groups according to the similarity of the survival curves; divide the employee data set into multiple data groups corresponding to the curve groups; use the survival analysis algorithm to calculate multiple survival rates corresponding to multiple time points in the data groups to obtain complex numbers corresponding to the curve groups respectively Survival curves after several clusters, and at least one cut-off time point is obtained according to the slope changes of the survival curves after clustering; and based on the at least one cut-off time point, a proportional hazards model is used to calculate the prediction result of turnover rate of the employees corresponding to the influencing variable. The electronic device as claimed in claim 7, wherein the processor is further configured to: divide a plurality of intervals based on the value range corresponding to the influencing variable; and divide the employee data set into the subsets corresponding to the intervals. The electronic device according to claim 7, wherein the processor is further configured to: use a moving average method to smooth the survival curves after grouping; and find at least one time point with the largest slope change from the smoothed survival curves after grouping as the at least one division time point. The electronic device as claimed in item 7, wherein the processor is further used for: Using the proportional hazards model to calculate the risk ratios of the curve groups in multiple time ranges, wherein the time ranges are set based on the at least slicing time point; using one of the curve groups as a baseline, comparing the risk ratios corresponding to the other curve groups with the risk ratio corresponding to the baseline; and using the comparison results as the turnover rate prediction results.

Images