Rolling mill drive bearings operate under extremely severe conditions, including high loads, elevated temperatures, and intense vibration, which leads to accelerated wear and premature failure. In industrial practice, the technical condition of such bearings is typically monitored indirectly using temperature and vibration measurements. However, these approaches do not ensure reliable prevention of emergency failures, as evidenced by multiple incidents in metallurgical plants caused by bearing overheating and damage. To address these limitations, this paper proposes a diagnostic approach based on vibration analysis combined with machine learning techniques. A diagnostic model using the Random Forest algorithm is developed to classify the technical condition of sliding bearings into three states: normal, warning, and critical. The model utilizes a feature vector composed of vibration and operational parameters, including RMS vibration, peak vibration, standard deviation, dominant frequency, temperature, load, rotational speed, and operating time since last maintenance. The model was preliminarily evaluated using a numerically generated dataset based on industrial measurements, rolling mill technical specifications, and established diagnostic threshold values. The obtained results should be considered as a preliminary assessment of the proposed approach and require further validation using extended real industrial datasets.