Deep Learning Framework for Employee Attrition Prediction and Performance Trend Analysis in Data-Driven HR Environments

The problem of employee attrition and performance trends have a critical effect on the HR strategy and organizational productivity. The current methods of predicting are usually based on single-model techniques, which restricts the capabilities of such methods to reproduce both time dynamics of performance and complicated organizational interactions. This paper suggests a hybrid deep learning model that combines a Time-Series Transformer to capture the temporal performance data, a DNN to capture the tabular HR variables, and a Graph Attention Network (GAT) to capture the inter-employee and team relationship to be able to predict the attrition and the performance trend at the same time. Using IBM HR Analytics Employee Attrition Dataset and Enterprise HR Performance Dataset the experiments were performed. The method was coded in Python using TensorFlow and Python Torch. The findings showed better accuracy, F1-score, and low RMSE than baseline models and interpretable results of SHAP and LIME analyses of important factors that affected attrition and performance. The framework provides a well-explainable, solid solution to data driven HR decisions, active management of the workforce and strategic planning.