The transient electromagnetic method is characterized by strong detection capability and broad applicability, making it widely used in the exploration of coal mine goafs. However, the complexity of data inversion and the high uncertainty of results limit its accuracy in identifying goaf types. To address this issue, Donggou coal mine was selected as the study area. Based on smoke ring inversion and depth correction, four TEM-derived parameters sensitive to the electrical response of goafs were chosen. Fifty measurement points were selected as training samples from known dry goafs, water-filled goafs, and normal strata, respectively, to establish a goaf type discrimination model using the Bayesian algorithm. The model achieved an accuracy of 84.7% in back-estimation and was successfully applied to classify the goaf types of measurement points across the entire 3^-3 coal seam. The resulting distribution characteristics were consistent with drilling verification. The study demonstrates that the Bayes discrimination model, based on TEM multi-parameter analysis, can quickly and accurately identify goaf types, effectively resolving the multi-solution problem of data inversion and significantly improving the reliability of TEM in coal mine goaf detection.