Diagnosis of Mental Health from Social Networking Posts: An Improved ML-Based Approach

Social networking and microblogging websites have rich information about user’s personal life and mental health. A systematic analysis of this information can be used to understand the mental and psychological state of the users. This can also be used for preventive decision-making in case of mental illness. Various machine learning techniques have already been used in past works to extract the user’s mental health from social media data. There is still a need to identify the most effective approach among the various machine learning techniques to diagnose sadness in the data that reflect negativity. With this motivation, we use five supervised machine learning algorithms with a hybrid of text feature extraction techniques: Term Frequency-Inverse Document Frequency (TF-IDF) and Bag of Words (BoW). Our results establish improved accuracy and precision than the state-of-the-art works. Experimental results show the usage of TF-IDF and BoW collectively with all the ML algorithms. It has been inferred that the support vector machine performs best among the various machine learning algorithms with the highest balanced accuracy of 99.7%.