Objective To investigate the electroencephalogram (EEG) characteristics of anxiety disorder patients and high-risk populations, and to provide Objective support for the military psychological selection and the theory of multi-quality integration. Methods The EEG data from anxiety disorder patients (anxiety disorder group, n＝38), individuals at high risk for anxiety disorders (anxiety high-risk group, n＝39), and healthy volunteers (normal group, n＝38) were collected and analyzed by the eeglab software to investigate the multi-quality integration indicators (power spectral density ［PSD］, time-frequency amplitude, and functional connectivity). Whether PSD and weighted phase lag index (wPLI) could serve as neurophysiological indicators for assessing anxiety disorder-related brain abnormalities were examined. Binary classification among the 3 groups was performed by support vector machine and K-nearest neighbors (KNN) classifiers from the Python 2.0 Scikit-Learn package. Results There were significant differences in the PSD among the 3 groups in the δ, θ, and α low-frequency bands. The main effect of the δ band between groups was significant (F＝97.55, P＜0.001), with the PSD of the anxiety disorder group (6.16±0.61)＞the high-risk anxiety group (5.22±0.73)＞the normal group (3.36±0.06). The main effect of the θ band between groups was significant (F＝65.87, P＜0.001), with the PSD of the anxiety disorder group (2.25±0.07)＞the high-risk anxiety group (2.23±0.08)＞the normal group (1.34±0.39). The main effect of the α band between groups was significant (F＝178.73, P＜0.001), with the PSD of the anxiety disorder group (2.02±0.45)＞the high-risk anxiety group (1.94±0.57)＞the normal group (0.98±0.02). In the β1, β2, and γ high-frequency bands, there were fluctuating increases in PSD in the prefrontal (FP1, FP2) and temporal (T₃, T4) regions of the high-risk anxiety group. In the β1 band, the differences in wPLI between the anxiety disorder group and normal group were significant at the TP7-FC3 (t＝2.45, P＜0.05) and T₅-FC3 electrode pairs (t＝－3.01, P＜0.05). By integrating behavioral, frequency domain, time-frequency, and functional connectivity features into machine learning, the multi-quality integration indicators improved the accuracy of identifying high-risk anxiety individuals from normal individuals from 75.00% to 82.61%, compared to using behavioral indicators alone. Conclusion Combination of a multi-quality integration approach with machine learning can identify distinguishing features of individuals at risk for anxiety disorders, which holds potential implications for military psychological selection and clinical assessments.