A single-channel algorithm was proposed in order to study effect of intermittent hypoxic training on hypoxia tolerance based on EEG pattern. EEG was decomposed by ensemble empirical mode decomposition into a finite number of intrinsic mode functions (IMFs) based on the intrinsic local characteristic time scale. Analytic amplitude, analytic frequency, and recurrence property quantified by recurrence quantification analysis were explored on IMFs, and the first two scales revealed difference between normal EEG and hypoxia EEG. Classification accuracy of hypoxia EEG and normal EEG could reach 67.8% before decline of neurobehavioral ability, which represented that hypoxia EEG pattern could be detected at an early stage. Classification accuracy of hypoxia EEG and normal EEG increased with time and deepened intensity of hypoxia was observed by regular shift of hypoxia EEG pattern with time in a three dimensional subspace. The reduced shift and classification accuracy after intermittent hypoxic training represented that hypoxia tolerance enhanced.