Musculoskeletal disorders (MSDs) are increasingly prevalent and a leading cause of disability worldwide. Injured workers with MSDs often undergo treatment at rehabilitation centers before returning to work. Functional capacity evaluation (FCE) is commonly used at these centers to assess an individual's physical ability to perform specific job tasks and readiness to return-to-work. However, one limitation of FCE is that the therapist's judgments may be influenced by information other than visual impressions. To address this, this study proposes a method for automatically categorizing work level categories to enhance traditional FCE. Male and female volunteers with and without previous MSDs were recruited to perform a core-lifting task for EMG activity analysis of the biceps brachii and erector spinae muscles. Spectrogram and S-transform were used for pre-processing and feature extraction. Machine learning techniques, including linear discriminant analysis (LDA), support vector machine (SVM), k-nearest neighbor (KNN), and artificial neural network (ANN), were utilized to accurately distinguish between medium and high work levels. LDA had the highest accuracy for training, validation, and testing, except for the testing accuracy of the individual right biceps brachii muscle, which was 90.9%.