The age and gender information extracted from Electroencephalogram (EEG) has been used in various applications which are allocating a person to age and gender groups, identifying or authenticating a person and improving brain-computer interface systems. Besides this, the EEG-based automatic recognition of alcoholics greatly supports to the psychiatrists. However, one of the major challenges when using EEG is about storing and transmitting a huge amount of EEG data, leading to the need of using compression. Although lossy compression techniques give much higher compression ratio (CR) than lossless ones, they introduce the loss of information including the age, gender and alcoholic information in the reconstructed signals, which may reduce the performance of EEG-based age, gender and alcoholic recognition systems significantly. In this paper, the impact of lossy compression on the age, gender and alcoholic information extracted from EEG signals is examined in detail with different feature extraction and machine learning techniques. Our experimental results indicate that with an appropriate feature extraction technique, we could minimize the information loss in EEG compression and maintain the high performance of EEG-based age, gender and alcoholics recognition systems.