PT Journal
AU Matejka, M
   Dian, M
   Lhota, J
   Beran, T
   Hlinak, V
TI Machine Learning Regression Approaches for Manufacturing Cost and Time Prediction: A Comprehensive Review
SO Manufacturing Technology Journal
PY 2026
BP 53
EP 62
VL 26
IS 1
DI 10.21062/mft.2026.010
DE Machine Learning; Regression; Cost Estimation; Time Prediction; Manufacturing
AB Today, machine learning regression methods are quietly but fundamentally transforming cost and time estimation in manufacturing: from early pricing to labor planning to operational order management. This survey offers a comprehensive map of approaches - from linear models, to tree ensembles (RF, GBM, XGBoost) and shallow neural networks, to multi-target and tensor regressions that can exploit data structure across BOM items and sequences of operations. With an emphasis on SME conditions, we show how to reconcile three often conflicting requirements of practice: accuracy, explainability, and integration into existing data flows (MES/ERP). The paper presents a comparative taxonomy of methods, recommended validation practices (MAE, RMSE, MAPE, R2  including confidence intervals) and a pragmatic adoption trajectory: from regularized multiple regressions to tree models to multi-output formulations sharing re-presentations across operations. Consolidated findings show that modern learners consistently outperform traditional baselines when supported by careful flag engineering, drift management, and data standardization. As a major research-application contribution, we propose a unified multi-objective framework for simultaneous cost and time prediction that combines domain (queueing/simulation) features with data-driven regression to enable transparent decision making in pricing and capacity planning. The study thus creates a bridge between theory and manufacturing practice and invites the reader to systematically but achievably deploy ML in everyday decision making.
ER