The Application of Supervised Machine Learning Algorithms for Improving the Accuracy of Manufacturing Operation Duration Prediction

Abstract

The study presents an approach to predicting the completion time of production operations using supervised machine learning techniques. The analysis was conducted on a database extracted from ERP and MES systems, comprising over 150,000 records containing technological and production completion times, operation numbers, and textual descriptions of operations. The dataset preprocessing involved cleaning, feature encoding, and text vectorisation using the TF-IDF method to represent semantic patterns within operation descriptions. Regression models, including Linear Regression, Random Forest, and XGBoost, were trained and evaluated using Google Colab. Model performance was assessed using standard evaluation metrics: Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and the coefficient of determination (R2). Experimental results showed that ensemble-based methods achieved the highest predictive accuracy, outperforming the baseline model based solely on technological completion time. In addition, the study examines the sensitivity of selected models to hyperparameter settings and analyses the impact of alternative categorical feature encoding methods on prediction accuracy. The proposed approach enables more accurate estimation of production durations and supports data-driven decision-making in manufacturing environments.