Analysis of Various Machine Learning Models for Detecting Depression in Twitter Tweets

Abstract

The main objective of this research paper is to help in identifying particular individual as depressive or not. In this research paper we have analysed and compared various machine learning approaches in comparison to our baseline machine learning approach Logistic Regression to identify a tweet as depressive or non-depressive. The tweets sentences are needed to be converted in to the form such as they could be fed to the machine learning models. Compared the two techniques used for this TF-IDF and Count Vectorizer. Other machine learning approaches thathave been analysed are Ridge Classifier, Multinomial Naive Bayes, Complement Naive Bayes, Stochastic Gradient Descent (SGD), Passive Aggressive Algorithms, Support Vector Classification, Voting classifier and Multi-layer Perceptron classifier. Voting classifier performed the best from all of them by giving approximately 80% of accuracy.