Abstract
In selecting promotion candidates, sometimes a manager can be in a difficultposition because he has to choose one of the many teams and their members. This
large selection of candidates could potentially lead to subjectivity. The intervention
needed to help identify promotion candidates can be carried out using classification
machine learning. This research presents the results of testing, evaluating and
improving several classification algorithms, namely: decision tree, deep learning,
K-nearest neighbor and random forest in predicting promotion candidates based
on multi-year performance assessments. The total sample was 230 employees with
a distribution of 216 middle management level employees and 14 senior
management employees. Evaluation of the performance of the classification
algorithm uses a confusion matrix, where in this research recall is prioritized over
accuracy and precision. Recall was chosen because the predicted results of a
promotion but not actually a promotion were more unexpected than the predicted
results that were not a promotion but actually a promotion. The research results
found that (1) the decision tree algorithm was a better algorithm among the others,
(2) the decision tree algorithm using principal component analysis (PCA) was
better than the one using the correlation matrix, combining PCA and correlation
matrix and without using both PCA and correlation matrix, (3) the variables age,
management level, and gender (male) become the three most influential variables
on predictions of job promotion candidates, and (4) the variables age, management
level and 2019 competency score become the three the most influential variable on
the prediction of grade promotion candidates.
| Date of Award | 2024 |
|---|---|
| Original language | Indonesian |
| Supervisor | Faisal Fahmi (Supervisor) |