Construction of a Machine Learning-Based Model for Predicting the Mechanical Properties of $\text{Fe}-\mathrm{C}-\text{Mn}-\text{Al}$ System Steels

In order to improve the prediction accuracy of mechanical properties of Fe-C-Mn-Al steel, XGBoost, LightGBM and deep neural network (DNN) models are constructed based on the experimental data, and feature engineering optimization, hyper-parameter tuning, and confidence-weighted integration strategies are adopted to improve the generalization ability of the models. The effects of different compositions and process parameters on yield strength, tensile strength, elongation and hardness are analyzed, and the results show that the multi-model fusion method can effectively reduce the prediction error, maintain high stability in industrial validation, and provide data support for the optimization of steel properties and the design of process parameters.