An optimized machine-learning tool to predict heat treatment response of hot-work tool steels

• Ensemble ML models achieve higher accuracy and lower errors than other models. • CatBoost & stacking models achieve R² values of 0.99 for HRC, 0.94 for K IC . • Tempering & hardening drive property prediction majorly over other inputs. • GUI by ML offers real-time predictions with impressive accuracy & usability. Hot-work tool steels are essential materials in high-performance tooling applications due to superior thermal fatigue resistance, strength, and toughness at elevated temperatures. Accurate prediction of mechanical properties, particularly fracture toughness (K IC ) and hardness (HRC), is crucial for their efficient design. However, conventional methods like tempering diagrams are often time-consuming, expensive, and lack broad applicability. This study proposes an optimized machine-learning (ML) framework for predicting HRC and K IC values of hot-work tool steels based on 2.564 experimental data points involving 13 input parameters, including chemical composition, processing routes, and heat treatment conditions. A comparative analysis of ten ML algorithms-ranging from decision trees and k-nearest neighbors to ensemble methods-was conducted. The originality of this work lies in the integration of advanced ensemble learning, SHAP-based model interpretation, and the creation of a graphical user interface (GUI) for real-time property prediction. Models were validated using both conventional test/train splits and 5-fold cross-validation, with their performance evaluated using mean squared error (MSE), mean absolute error (MAE), root mean squared error (RMSE), mean absolute percentage error (MAPE), and R-squared. CatBoost and stacking achieved highest accuracy of 99% for HRC and 94% for K IC , with mean absolute errors of 0.3 HRC and 3.3 MPa.m -1/2 , respectively. Additionally, the GUI tool, which integrates machine learning-derived empirical equations, showed impressive prediction accuracy (R 2 = 99% for HRC and 95% for K IC ), providing a valuable resource for materials engineers. This research presents a strong, data-driven approach to accelerate the development of tool steel while reducing the need for extensive experimental procedures.