A KNOWLEDGE GRAPH-BASED GRAPH CONVOLUTIONAL NETWORK MODEL FOR EMPLOYEE PERFORMANCE PREDICTION

ABSTRACT

The rapid growth of organizational data has increased the need for computational approaches capable of forecasting employee performance from both individual attributes and organizational relationships. This study develops and evaluates a knowledge graph-based Graph Convolutional Network (GCN) model for employee performance prediction. The research used 1,000 anonymized employee records selected from a larger Kaggle repository described in the source study. Employee attributes were cleaned, encoded, standardized, and represented as graph nodes with relational links based mainly on organizational characteristics such as department and job role. A two-layer GCN implemented with graph learning tools was trained and evaluated under several train-validation-test configurations. The study used accuracy, precision, recall, F1-score, confusion matrices, training curves, and GNNExplainer-based explanations to assess predictive performance and model transparency. The baseline experiment produced 18.67% accuracy, 18.22% precision, and 17.17% F1-score. Across alternative data splits, accuracy ranged from 17.33% to 24.50%. The highest accuracy was recorded under the 65-15-20 split, while the highest precision was 31.00% under the same configuration. The results show limited predictive strength, especially for minority performance categories, but they also demonstrate the value of relational representations for exposing organizational patterns. GNNExplainer repeatedly identified job title, education level, department, and department-centered structural relationships as important explanatory elements. The study therefore establishes a reproducible baseline for knowledge graph-based HR analytics and identifies class imbalance, limited graph richness, small experimental sample size, and shallow model architecture as major areas for improvement.

Keywords: Employee Performance Prediction, Knowledge Graph, Graph Convolutional Network, Graph Neural Network, Explainable Artificial Intelligence, Human Resource Analytics, Workforce Analytics