Performance Evaluation of Learning Models for Identification of Suicidal Thoughts

Abstract

Abstract The suicidal death rate is growing rapidly. Depression and stress levels among the people have increased significantly, which is considered to be a risk factor for suicidal thoughts. Social media is gradually more popular and people use them for sharing their sentiments and harmful emotions related to suicidal thoughts. An effective approach is required to investigate for identifying risk factors associated with suicide on social media. The objective is to propose some learning models to evaluate social media data to identify persons having suicidal tendencies. A large data consisting of 8452 tweets are collected from Twitter, pre-processed and bags of words were applied. Different machine learning and deep learning algorithms such as Random Forest, Decision Tree, Bernoulli Naïve Bayes, Multinomial Naïve Bayes, Recurrent Neural Network, Artificial Neural Network and Long Short Term Memory were applied for classifying the tweets in two sets: suicidal and non-suicidal. The performance of these learning models is further evaluated on three parameters: accuracy, precision and recall. These models have shown significant results on the parameters.