Comparative Machine Learning Analysis for Employee Performance Prediction: Production-Ready Java EE System with Novel BLOB-Based Persistence

Inefficient employee performance management systems, which depend on subjective evaluations and lack the ability to predict future performance, cost corporations millions each year. This study details the design, execution, and rollout of an enterprise-level Employee Performance Analytics and Prediction System built on Java EE 8. This system fills important gaps between machine learning research and real-world application. Using a dataset with 17,417 employee records from Kaggle's HR Analytics repository, we tested three classification algorithm: J48 Decision Tree, Random Forest, and Naive Bayes using the Weka 3.8.6 machine learning library. Our empirical evaluation through 3-fold stratified crossvalidation demonstrates that Naive Bayes achieves optimal performance with 81.15% accuracy, outperforming ensemble methods. Atomic transactions, SQL-based model selection, and cluster-wide accessibility are made possible by the system's innovative BLOB-based model persistence mechanism, which addresses the stateless session problem present in enterprise Java environments. The system, which is deployed on a WildFly 18.0.1 application server with dual web service interfaces (JAX-RS 2.1 RESTful and JAX-WS 2.2 SOAP), exhibits production-grade performance: 142ms average prediction latency, sustained throughput of 35 predictions/second under concurrent load, and 99.7% uptime during 72-hour stress testing with 50 concurrent users executing 2,834 requests. This complete implementation demonstrates that enterprise Java EE platforms are still practical for contemporary machine learning deployments while giving businesses a standardsbased predictive analytics infrastructure.