Table 1
Attributes in the first dataset.
| Attribute | Description |
|---|---|
| ID | Staff ID |
| Gender | Gender of the staff |
| Marital status | Marital status of the staff |
| Birth Date | Birth date of the staff |
| Education level | Education level of the staff |
| Graduated School | School the staff graduated from |
| Graduated Department | Department the staff graduated from |
| Department | Department of staff in the company |
| Duty | Duty of the staff in the company |
| Total Working Years | Total years of experience of the staff |
| Work Experience 1 - Working Time | Years of work experience in the first company |
| Work Experience 2 - Working Time | Years of work experience in the second company |
| Work Experience 3 - Working Time | Years of work experience in the third company |
| Universal Entry Date | Entry date of the staff to Universal |
| Current Salary | Current salary of the staff |
| Weekly Working Hours | Hours per week the staff works |
| Job Satisfaction | Staff satisfaction score with his job |
| Internal Relationship Satisfaction | Internal relationship satisfaction score of the staff |
| Business Travel Frequency | The number of business trips of the staff |
| Address - District | District where the staff lives |
Table 2
Attributes in the second dataset.
| Attribute | Description |
|---|---|
| Staff Type | Type of the staff |
| Staff Subtype | Subtype of the staff |
| Gender | Gender of the staff |
| Marital Status | Marital status of the staff |
| Agreement | Agreement between company and staff |
| Education level | Education level of the staff |
| Department 1 | The department to which the staff is affiliated |
| Department 2 | The department where the staff works |
| Insured Unit | Insurance policy of the staff |
| Duty | The duty assigned to the staff |
| Address - Province | Province where the staff lives |
| Address - District | District where the staff lives |
| Age | Age of the staff |
Table 3
Parameter values staff churn prediction model developed using LR, CatBoost and ELM.
| Algorithm | Parameters | Values |
|---|---|---|
| LR | Penalty | 12 |
| LR | C | 105 |
| CatBoost | N_estimators | 100 |
| CatBoost | Max_depth | 6 |
| CatBoost | Iterations | 500 |
| CatBoost | Learning_rate | 0.03 |
| ELM | Alpha | 10−5 |
| ELM | N neurons | 64 |
| ELM | Activation Function | tanh |
Fig. 1
Complexity matrix obtained with LR.
Fig. 2
Complexity matrix obtained with Catboost.
Fig. 3
Complexity matrix obtained with ELM.
Fig. 4
Complexity matrix obtained with LR.
Fig. 5
Complexity matrix obtained with Catboost.
Fig. 6
Complexity matrix obtained with ELM.
Fig. 7
Complexity matrix obtained with LR.
Fig. 8
Complexity matrix obtained with Catboost.
Fig. 9
Complexity matrix obtained with ELM.
Table 4
Accuracy and F1-score values of staff churn prediction models developed using LR, Catboost and ELM.
| Approaches | Algorithms | F1-Score | Accuracy |
|---|---|---|---|
| First Approach | LR | 0.68 | 0.77 |
| First Approach | CatBoost | 0.94 | 0.95 |
| First Approach | ELM | 0.66 | 0.72 |
| Second Approach | LR | 0.67 | 0.76 |
| Second Approach | CatBoost | 0.91 | 0.93 |
| Second Approach | ELM | 0.65 | 0.70 |
| Third Approach | LR | 0.60 | 0.72 |
| Third Approach | CatBoost | 0.92 | 0.94 |
| Third Approach | ELM | 0.65 | 0.71 |
Table 5
Parameter values of staff lifetime prediction model developed using SVM, LightGBM and KNN.
| Methods | Parameters | Values |
|---|---|---|
| SVM | C | 1010 |
| SVM | Kernel | RBF |
| SVM | Gamma | Scaled |
| LightGBM | N_estimators | 50 |
| LightGBM | Num_leaves | 30 |
| LightGBM | Max_depth | None |
| LightGBM | Learning_rate | 0.1 |
| KNN | N_neighbours | 5 |
Fig. 10
Predicted values and actual values generated by SVM.
Fig. 11
Predicted values and actual values generated by LightGBM.
Fig. 12
Predicted values and actual values generated by KNN.
Fig. 13
Predicted values and actual values generated by SVM.
Fig. 14
Predicted values and actual values generated by LightGBM.
Fig. 15
Predicted values and actual values generated by KNN.
Fig. 16
Predicted values and actual values generated by SVM.
Fig. 17
Predicted values and actual values generated by LightGBM.
Fig. 18
Predicted values and actual values generated by KNN.
Table 6
Accuracy and F1-score values of staff churn prediction models developed using LR, Catboost and ELM.
| Approaches | Algorithms | MAE | MAPE (%) |
|---|---|---|---|
| First Approach | SVM | 55.44 | 7.24 |
| First Approach | LightGBM | 87.97 | 10.44 |
| First Approach | KNN | 83.33 | 10.95 |
| Second Approach | SVM | 76.54 | 8.69 |
| Second Approach | LightGBM | 102.86 | 11.37 |
| Second Approach | KNN | 99.53 | 12.12 |
| Third Approach | SVM | 62.70 | 8.01 |
| Third Approach | LightGBM | 83.98 | 8.41 |
| Third Approach | KNN | 83.33 | 10.95 |
Table 7
Parameter values of staff churn prediction models.
| Algorithms | Parameters | Values |
|---|---|---|
| LR | Penalty | 12 |
| LR | C | 105 |
| CatBoost | N_estimators | 100 |
| CatBoost | Max_depth | 6 |
| CatBoost | Iterations | 1000 |
| CatBoost | Learning_rate | 0.1 |
| ELM | Alpha | 10−3 |
| ELM | N neurons | 64.32 |
| ELM | Activation function | tanh |
Fig. 19
Complexity matrix obtained with LR.
Fig. 20
Complexity matrix obtained with Catboost.
Fig. 21
Complexity matrix obtained with ELM.
Fig. 22
Complexity matrix obtained with LR.
Fig. 23
Complexity matrix obtained with Catboost.
Fig. 24
Complexity matrix obtained with ELM.
Fig. 25
Complexity matrix obtained with LR.
Fig. 26
Complexity matrix obtained with Catboost.
Fig. 27
Complexity matrix obtained with ELM.
Table 8
Accuracy and F1-score values of staff churn prediction models developed using LR, Catboost and ELM.
| Approaches | Algorithms | F1-Score | Accuracy |
|---|---|---|---|
| First Approach | LR | 0.85 | 0.85 |
| First Approach | CatBoost | 0.91 | 0.91 |
| First Approach | ELM | 0.87 | 0.86 |
| Second Approach | LR | 0.88 | 0.87 |
| Second Approach | CatBoost | 0.90 | 0.89 |
| Second Approach | ELM | 0.83 | 0.82 |
| Third Approach | LR | 0.78 | 0.77 |
| Third Approach | CatBoost | 0.91 | 0.90 |
| Third Approach | ELM | 0.76 | 0.75 |
Table 9
Parameter values of staff lifetime prediction models.
| Methods | Parameters | Values |
|---|---|---|
| SVM | C | 105 |
| SVM | Kernel | RBF |
| SVM | Gamma | Scaled |
| LightGBM | N_estimators | 100 |
| LightGBM | Num_leaves | 50 |
| LightGBM | Max_depth | None |
| LightGBM | Learning_rate | 0.1 |
| KNN | N_neighbours | 5 |
Fig. 28
Predicted values and actual values generated by SVM.
Fig. 29
Predicted values and actual values generated by LightGBM.
Fig. 30
Predicted values and actual values generated by KNN.
Fig. 31
Predicted values and actual values generated by SVM.
Fig. 32
Predicted values and actual values generated by LightGBM.
Fig. 33
Predicted values and actual values generated by KNN.
Fig. 34
Predicted values and actual values generated by SVM.
Fig. 35
Predicted values and actual values generated by LightGBM.
Fig. 36
Predicted values and actual values generated by KNN.
Table 10
Accuracy and F1-score values of staff churn prediction models developed using LR, Catboost and ELM.
| Approaches | Algorithms | MAE | MAPE (%) |
|---|---|---|---|
| First Approach | SVM | 1158.25 | 19.94 |
| First Approach | LightGBM | 1011.81 | 18.37 |
| First Approach | KNN | 999.96 | 18.35 |
| Second Approach | SVM | 1124.01 | 19.68 |
| Second Approach | LightGBM | 1038.58 | 19.03 |
| Second Approach | KNN | 1044.15 | 18.89 |
| Third Approach | SVM | 1087.65 | 19.07 |
| Third Approach | LightGBM | 1045.42 | 18.85 |
| Third Approach | KNN | 1068.99 | 19.19 |