Machine Learning‐Based Regression Modeling Approach to Predict the Carbon Content in Molten Steel of Electric Arc Furnace

This research presents the development of a decision‐support system for predicting the carbon content using data‐driven modeling to reduce carbon emissions and power consumption during the electric arc furnace process. The decision‐support system is developed in the following sequence: 1) collect raw data from the company and performs data preprocessing (e.g., outlier detection) using statistical methods to obtain refined data, 2) select features based on domain knowledge shared by communicating with field experts, 3) develop a machine learning‐based regression model, divided into low‐carbon steel group and high‐carbon steel group, and 4) identify correlations between input variables and predicted output values with sensitivity analysis. The results show that the predicted carbon contents are within a 0.05% margin of error compared to the actual values for 99% of the low‐carbon steel data and 97% of the high‐carbon steel data. Furthermore, it indicates that the carbon and oxygen are added after the first temperature measurement, and the carbon content at the first temperature measurement has the greatest impact on the final carbon content. The outcome of this research can contribute several practical benefits, such as reducing electricity consumption and shortening operation time by minimizing the number of sampling procedures.