This study aims to realize a seizure detector using the empirical mode decomposition (EMD) algorithm and a machine learning–based classifier that is robust enough for practical applications. We conducted exhaustive tests on electroencephalography (EEG) data of 24 pediatric patients who suffered from intractable seizures. We used the mean frequency of the first and the last intrinsic mode function (IMF) components within each two-second seizure epoch as our primary feature, which we tested on a total of 933 hours of data and 193 seizures spanning 3.11 hours. We obtained encouraging results for 10 patients, with average value of experimental metrics such as detection accuracy (sensitivity) of 100%, specificity of 95.5%, and latency of 2.53 seconds. These figures are competitive as compared to similar recent studies and suggest that the proposed method is helpful for some epilepsy patients. As the method has relatively low computational complexity, it may serve as a potential avenue for real-time seizure detection.