Employee Efficiency Measurement Using Elastic Net Regression in Enterprise Resource Performance Systems

The evaluation of people in working conditions is complicated by a dilemma of structural analysis due to the emergence of new Enterprise Resource Planning (ERP) software, which provides huge amounts of different performance measurements of employees, covering their behavior, work performance, and degree of engagement on the monthly basis by providing signals from workers. Much of this valuable information turns out to be redundant, which means that it is correlated with other types of signals, and only a small number of them can help in practicing successful predictions of the performance outcome. Traditional analytical techniques, in particular Ordinary Least Squares regression (OLS), which are used for analyzing an overall data pool from all the sources, fail at two levels-they are subject to overfitting of correlated variables (having large coefficients' variations) and provide unstable models, which are very hard to interpret and thus use by HR managers. Elastic Net regression, which combines L1 (LASSO) and L2 (Ridge) penalties in a single objective function, addresses both failure modes: the L1 penalty induces sparsity (eliminating redundant features), and the L2 penalty stabilises coefficient estimates under multicollinearity (regularising correlated features toward each other). This paper applies Elastic Net regression to employee efficiency score prediction across a dataset of 18,400 employee records from six enterprise organisations (manufacturing: 2 firms, financial services: 2 firms, technology: 2 firms) spanning 24 months (January 2023-December 2024). Forty-eight performance features across seven behavioural and operational categories are extracted from ERP system logs, HR process records, and learning and development platforms. With mixing parameter${\alpha}=\mathbf{0. 6 4}$(L1 fraction) and regularisation strength${\lambda}=\mathbf{0. 0 2 4}$, selected by 5-fold cross-validation, Elastic Net selects 24 of 48 features (50% sparsity) and achieves RMSE$=\mathbf{6. 8 4}$(on a$\mathbf{0}-\mathbf{1 0 0}$efficiency scale) and$\mathbf{R}^{\mathbf{2}}=\mathbf{0. 8 8 4}$on the held-out test set$(\mathbf{n}=\mathbf{1, 8 4 0}; \mathbf{1 0 {\%}}$stratified split)-outperforming OLS$(\mathbf{R}^{\mathbf{2}}: \mathbf{0. 6 2 4) }$, LASSO (0.784), Ridge (0.824), and approaching Random Forest (0.904) while retaining full coefficient interpretability. The five dominant efficiency predictors by standardised coefficient magnitude are task completion rate$({\beta}=\mathbf{0. 2 4 8})$, ERP login sessions per day (0.184), training completion rate (0.168), on-time delivery rate (0.148), and quality error rate$({\beta}=-\mathbf{0. 1 2 8})$. Deployed as a monthly efficiency scoring pipeline, the Elastic Net model enables targeted bottom-quartile interventions that achieve a 24.8% performance improvement over a 12-month window and a promotion recommendation accuracy of 84.6%.

Hali tarjima qilinmagan