A Hybrid Feature Extraction and Classification Framework for Depression Detection using Deep Learning and Machine Learning Techniques

Depression is most common mental disorder that is affecting approximately 280 million individuals in the world. The stigma and lack of acceptance and awareness is still influencing people and that is making its diagnosis challenging for the medical practitioners. For tackling this situation technologies are being used in various domains. Speech processing is one such domain that is being used for detection of depression. In that, detection severity of depression with high accuracy remains an active area of research. The work presented in the paper provides an efficient method for feature extraction by spectrogram based CNN, feature selection by a two-level approach of ANOVA and ABC optimizer and multi-level classification by SVM (linear). This hybrid framework provides classification accuracy of 80% which validates its potential for application in real-world scenarios.