Lifting electromagnets are widely used for attracting, holding, and transporting ferromagnetic loads. In addition to contact-based operation, in which the load is brought into direct contact with the electromagnet poles, such systems can also function as electromagnetic suspension devices, where the load is maintained at a specified air gap without mechanical contact with the poles. In this case, static admissible operation is defined by the balance between the load weight and the electromagnetic attractive force within allowable air-gap and force limits. This study proposes a machine learning-assisted framework for static stability-region classification and magnetomotive force prediction in the design-stage evaluation of a lifting electromagnet operating in suspension mode. A computational database was generated from the mathematical model of the electromagnet, where the air gap g, magnetomotive force F, and electromagnetic attractive force P_E were considered as the main design variables. The database contained 13992 calculated operating points, with g varying from 4.8×10^(-5) to 3.1605×10^(-3)m, P_E from 0.22301 to 341.715326 N, and F from 150 to 425 A·turns. The predefined static stability-region limits resulted in 448 stable and 13,544 unstable samples. Random Forest, Gradient Boosting, and Decision Tree classifiers were trained and compared using the area under the receiver operating characteristic curve (AUC) as the main metric. The Decision Tree Classifier achieved the best result for the considered dataset, with an AUC of 1.000 and the shortest training and prediction times. A regression problem was then formulated to predict the required magnetomotive force for specified air-gap and electromagnetic-force values. Random Forest, Gradient Boosting, multilayer perceptron, and CatBoost regressors were evaluated. The CatBoost Regressor achieved the highest overall prediction accuracy, with a coefficient of determination R2=0.999, root mean square error of 1.7933 A·turns, mean absolute error of 1.2746 A·turns, mean absolute percentage error of 0.4630%, and a prediction time of approximately 0.000023 s. The obtained results show that the proposed framework can serve as a design-stage surrogate tool for rapid static operating-region classification and magnetomotive force prediction.