Detecting Depression in Social Media using Machine Learning Algorithms

Social media platforms such as Facebook, Twitter, and Instagram drastically change our world. After the Covid-19 outbreak, people become more inclined to discuss their sentiments and emotions on social media. Every year, over 300 million individuals worldwide suffer from depression, a prevalent mental condition that can lead to suicide. Early detection of depression on social media can help save lives. The proposed work aims to identify depressive qualities in tweets. The main challenge of this task is performing feature engineering and modeling to classify tweets as depressed or not. Additionally, estimating both depressive and non-depressive traits in a tweet is essential for fully understanding its content, which is crucial for determining how best to assist the user. We propose a probabilistic machine learning (ML) approach that provides not only a binary classification but also a probability estimate for the classification. This approach allows us to capture the uncertainty and varying degrees of depressive traits in tweets, leading to more nuanced insights. The proposed model is tested using datasets generated from mining depressive tweets and an equal number of random tweets, totaling 39,111 tweets. By improving data collection with precise keywords and enhancing feature engineering through proper tweet cleaning, the proposed model’s accuracy increases by 3%. The probabilistic approach further enhances the model's ability to assess the likelihood of depressive characteristics.