Benchmarking AI-Driven Classification Approaches in Employee Performance Forecasting

Abstract

Introduction: Employee performance is a critical factor in achieving organizational goals. Human Resource (HR) departments often struggle with timely and objective performance assessments, which are essential for decisions like promotions, terminations, and training. This paper investigates the application of ML techniques to predict and categorize employee performance based on structured data. We benchmarked several supervised classification models—including DT, RF, NN, SVM and LR implemented via Altair AI Studio (RapidMiner). Models were evaluated using RMSE, MAE, and classification accuracy. Among them, the DT model demonstrated the highest accuracy and interpretability, making it ideal for HR applications.

Objectives: The objective of the paper is to design a ML-based predictive model to classify employee performance using structured data. By implementing and evaluating various supervised classification algorithms—including DT, RF, NN, SVM, and Logistic Regression—the aim is to classify the most effective model for predicting employee outcomes. Performance will be assessed using metrics such as RMSE, MAE, and classification accuracy. The paper seeks to provide HR departments with a trustworthy, data-driven tool to make informed decisions regarding promotions, terminations, and training, ultimately enhancing organizational performance and efficiency.

Methods: The methodology for this project consists of several essential steps. First, we collected structured employee data, including performance metrics, job roles, and demographic information. Next, we prepared the data by handling missing values, correcting categorical attributes, and normalizing continuous features. Various supervised classification algorithms, including DT, RF, Neural Networks, SVM, and Logistic Regression, were implemented using Altair AI Studio (RapidMiner). Model performance was evaluated through metrics like RMSE, MAE, and classification accuracy. Finally, we compared the models’ results to identify the most suitable one for accurately classifying employee performance, focusing on interpretability and accuracy.

Results: The results obtained show the performance of different models (RF, DT, Neural Net, and SVM) based on various metrics. The DT model surpasses the other models in both RMSE (0.397) and absolute error (0.149). It also shows the lowest relative errors across all categories: 6.97 (relative error), 5.44 (lenient), and 7.87 (strict). RF has a relatively high RMSE (0.96) and absolute error (0.739). Neural Net and SVM perform similarly, with higher RMSE and error values. The DT also predicts correctly for 24,032 out of 30,000 instances.

Conclusions: In conclusion, benchmarking AI-driven classification approaches in employee performance forecasting highlights the effectiveness of various ML algorithms. Among the models tested, RF demonstrated robust performance with high accuracy and interpretability, making it a strong contender for real-life applications. SVM showed promise in handling complex, high-dimensional data, while Deep Learning models excelled in capturing intricate patterns with large datasets, albeit at the cost of interpretability. Despite these strengths, challenges related to data quality, model explainability, and scalability persist. Upcoming research should focus on improving these aspects to improve the practical implementation of AI in employee performance forecasting.