Mental health analysis using deep learning for feature extraction

Deepali J. Joshi, Mohit Makhija, Yash Nabar, Ninad Nehete, Manasi S. Patwardhan

Abstract: There is an immense need to analyze and monitor a person's mental health as justified in our previous work. Feature extraction is a decisive part of all the data mining related tasks. We utilize deep learning feature extraction algorithm like sentence embedding to analyze mental health of persons from their social media posting and behavioral features and combine it with the traditional machine learning algorithms to enhance their performance. Newly, deep learning approaches have emerged as a novice way of constructing meaningful representations from unstructured data. Not only they are good in the data encoding but also carry the semantic meaning with them which help in modeling better. We find that deep learning feature extraction helps in classifying the normal users from the non-normal users as compared to their traditional counterparts. Also, the newer models attain a very low false positive rate. The best accuracy received by our model is 89%.