ENHANCING EMPLOYEE ATTRITION PREDICTION: THE IMPACT OF DATA PREPROCESSING ON MACHINE LEARNING MODEL PERFORMANCE

Abstract

Organizations face a serious problem with employee attrition, which raises expenses and reduces productivity. This study looks at how preprocessing data can help machine learning models forecast employee turnover more accurately. Seven machine learning algorithms—Random Forest, k-Nearest Neighbors (k-NN), XGBoost, Gradient Boosting, Linear Discriminant Analysis (LDA), LightGBM, and Logistic Regression—were used to analyze the 1,470 records in the International Business Machines Human Resources (IBM HR). Employee Attrition dataset. SimpleImputer was used to handle missing values, StandardScaler was used to standardize numerical features, and SelectFromModel was used to choose important features. These actions were essential in improving the accuracy of the model; LDA had the highest accuracy of 87.38%, followed by LightGBM and Logistic Regression, both of which had 87% accuracy. All models' performance metrics were much enhanced by preprocessing; k-NN had the lowest accuracy, at 85.33%. These results demonstrate how important preprocessing is to predictive analytics and how HR management may use it to identify at-risk workers and create successful retention plans.