The goal of this thesis was to gather, and study feature selection methods, and measure their capabilities, and performance, with special regards to those methods, which use deep neural networks. These, later mentioned methods include the Output Sensitivity analysis, the Signal to Noise Ratio analysis, the Activation Potential Analysis, and finally, Deep Feature Selection (DFS). The purpose of this task was to find a method, which is applicable to a healthcare dataset, that the department received from a collaboration with a foreign organization, a few years ago. This dataset contains information about the patients’ mental and physiological state, and some of their environmental variables, with the target feature being whether the patient had a clinically diagnosed depression in their lifetime. I’ve studied, and benchmarked the methods that I found, using two, artificially generated datasets, both of them having different similarities with the real data, and finally, I made a conclusion regarding which methods are the most capable for the analysis of the real dataset. In total, I think I’ve succeeded in selecting the two best, most capable methods, with results that support each other.