This study presents a machine learning-based approach for predicting the mechanical properties of ShKh15 (AISI 52100) bearing steel produced from secondary raw materials. A dataset comprising 180 experimental samples with varying chemical compositions, heat treatment parameters (austenitizing temperature 820–880°C, tempering temperature 150–250°C), and recycled material ratios (20–80%) was compiled from laboratory testing. Three ML models — Random Forest (RF), Artificial Neural Network (ANN), and Gradient Boosting Regression (GBR) — were trained and evaluated for predicting Rockwell hardness (HRC), tensile strength (σb), and impact toughness (KCU). The ANN model with architecture 9-64-32-3 achieved the best performance with R² = 0.9724 for hardness and RMSE = 0.87 HRC. Feature importance analysis identified recycled material ratio (31.4%), austenitizing temperature (22.8%), and carbon content deviation (18.6%) as the most influential parameters. The developed models enable rapid property estimation during bearing steel production from secondary materials, potentially reducing quality control testing by 45–55%.