Depression is the most common psychiatric disorder worldwide, which affects more than 300 million people. We aimed to detect depressed patients and healthy people automatically. We work on the PHQ-9 questionnaires and reduced it to a PHQ-5 questionnaires with a new cut-off value of 8 to detect depressed patients. We trained a Neural Network with 70% of our dataset. Then, the proposed classifier was tested with two datasets. The first one consists of 30% of PHQ-5 datasets, which could achieve 85.69%, 99.11% and 90.56% for accuracy, sensitivity and specificity respectively. The second test dataset consists of physical patient's parameters which recorded during a study in the Hanover Medical School. This classifier has shown good results in the detection of depression based on these two datasets.