EMPLOYEE ATTRITION PREDICTION USING MACHINE LEARNING ALGORITHMS
Lekan AKINODE & Olufunke BADA , August 2022, Published: 2022-09-20
Employees are the most important asset to every organisation. They play critical role in every organisation as they
help to generate innovative solutions for the overall development. Employee attrition is regarded as a well-known
issue that requires the administration to make the best choices in order to retain highly competent staff. It has indeed
been viewed as a crucial concern for businesses due to the negative impact it has on workplace motivation and
productivity. This research work compares supervised machine learning methods including Logistic Regression,
Naive Bayes Classifier, Random Forest Classifier, and XGBoost in order to forecast an employee's position within an
organization. Our research was evaluated using the IBM analytics unbalanced dataset (IBM ,2021) which includes
35 attributes for 1470 employees. Pre-processing was carried out to remove irrelevant features. We created a
balanced version from the original dataset in order to obtain results that were realistic. Models are evaluated using
evaluation metrics in order to find the most effective model given the available data. Hyperparameter optimization
was applied to further improve the results. Finally, cross-validation was used to properly assess our effort. The
software was developed using python and anaconda. After reviewing the data, it was discovered that the XGBoost
classifier has a better accuracy that is higher than the other techniques, at 85.5 percent, followed by the Random
Forest classifier at 84 percent. Precision is higher for the XGBoost classifier (0.72), then for the Random Forest
classifier (0.67). The decision tree classifier came in second with an F1 Score of 0.36, while the XGBoost classifier
achieved the best F1 Score of 0.40. The XGBoost classifier showed a greater performance in area under the curve
(AUC) with 0.50 in the ROC curve. The prediction analysis using the original data set for XGBoost was 85.5%,
however it was about 84.6% using a synthetic dataset.
Keywords: Attrition, Employee, Machine Learning, Algorithms


