Predictive Analytics for Mental Health Crises using Social Media Data with Attention Mechanism based Support Vector Machine Classification

Nowadays, the Mental Health Crisis (MHC) has increased into a demanding public health issue, requiring complete and extensive services. However, the number of people suffering from MHC is strictly not recorded, creating it more challenging to provide help and assistance. Hence, a hybrid model of Machine Learning (ML) namely Support Vector Machine along with a Self-Attention Mechanism (SVM-SAM) is proposed to identify individuals at high risk of mental health crises. Firstly, the data collected from Twitter API integration dataset, where the tweets contained urls, usernames, acronyms, and swear words. Next, the data is preprocessed by using Natural Language Tool Kit (NLTK), as it performs spelling correction. Then, features are extracted with TextBlob to extract the positive, negative, and neutral tweets of user by date. After that, Self-Attention Mechanism is utilized to select optimal features based on Dot-Attention Mechanism (DAM). Finally, the SVM classifier is introduced to classify the mental health crisis linearly separable based on user emotions into positive and negative sentiments. From the results, the proposed SVM-SAM performed better when compared with LSTM by providing results in terms of accuracy and specificity.