Abstract: Electromagnetic radiation (EMR) has proven to be an effective monitoring technology for coal rock dynamic disasters, including underground rock burst and coal and gas outbursts. However, the intricate generation mechanisms of electromagnetic signal, coupled with interference from underground environments, can compromise the accuracy of disaster monitoring and early warning systems. Accurately identifying EMR signals induced by coal rock fractures (effective signals) is essential for the widespread application of this technology. This study conducted monitoring experiments on electromagnetic radiation during uniaxial compression of coal rock, analyzing the time-domain, frequency-domain, and fractal characteristics of both valid and interference signals. Machine learning algorithms, such as linear discriminant analysis, support vector machines, and ensemble learning methods, were utilized to develop intelligent identification models for effective and interference signals. A comparative analysis of the recognition accuracy across different models was performed. The results demonstrated that characteristics like fractal box dimension, average frequency, count, and peak frequency effectively distinguished between valid and interference signals, with single-feature recognition accuracy surpassing 70%. Both the feature set and the choice of machine learning algorithm significantly influenced the identification accuracy of valid and interference signals. The ensemble learning method, leveraging the complete feature set, achieved the highest identification accuracy of 94.5% for both signal types, fulfilling the requirements for EMR monitoring and early warning applications.