Predicting Employee Attrition Using LSTM Networks And K-Means Clustering

Abstract

Staff turnover continues to be a major issue in Human Resource Management (HRM), which hurts how well a company performs and causes financial losses. It also messes up normal business activities. This research presents a new method that uses LSTM networks along with K-Means clustering to identify employees who are likely to leave their jobs. It also sorts employees into groups based on how likely they are to leave. The study uses the IBM HR Analytics dataset, which has information on 1,470 employees and includes 35 different factors like age, job history, and how happy they are at work. First, K-Means clustering is used to find employees who behave differently from others. Then, the LSTM model looks at patterns over time within each group. The performance of this LSTM-KMeans approach is checked using several important measures like Accuracy, Precision, Recall, F1 Score, and AUC-ROC. The results show that the LSTM-KMeans model does better than traditional methods like Support Vector Machine (SVM), Random Forest, and CatBoost in all of these areas, with average scores of 93.7%, 92.4%, 91.8%, and 92.1%, respectively. An ablation study also shows that the LSTM model with clustering performs better than the one without clustering on all the evaluation measures. These results highlight how combining unsupervised grouping with deep learning can give useful insights for HR and help create better strategies to keep employees.