Distress is often the cause of the global increase in cases of suicide. Consequently, in order to mitigate the belongings of depression, a clear diagnosis and course of treatment are necessary. An electroencephalogram (EEG) is used for recording and measuring the electrical impulses of the brain. It can be applied to produce a precise evaluation of depression severity. Prior studies have demonstrated the feasibility of diagnosing mental disorders employing deep learning (DL) models and EEG data. According to this research, the EEG data of individuals who are sad and those who are healthy are classified using a convolution neural network (CNN) built upon DL called DeprNet. In this case, the severity of the depression is indicated by the Patient Health Questionnaire 9 score. This paper presents the presentation of DeprNet in two experiment: the subject-wise divide and the record-wise split. The findings from the analysis of DeprNet demonstrate an efficiency of 0.9937 and an area below the receiver’s operating characteristic curve (AUC) of 0.999 while record-wise split data are included. Yet, an accurateness of 0.914 and an AUC of 0.956 can be achieved whenever subject-wise split data are employed. These consequences suggest that when CNN is programmed on record-wise split data, it overstrains on EEG data given a limited number of participants. DeprNet performs admirably when compared to one of the eight baseline models. Furthermore, it is found that in depressed people, the values of the ultimate CNN layer are dominant on the right electrode, while in typical participants, the values on the left electrodes are significant.