Depression Detection from Social Media Posts using Ensemble Machine Learning

Depression is a significant global health concern, especially among social media users who experience mental pressures. It is a vital mental health issue due to undiagnosed and untreated that can lead to chronic mental disease and even suicide. In the recent decade, depression has increased day by day. It needs early detection and treatment to improve people’s mental health. This paper proposed an ensemble machine learning approach to detect human depression from social media posts efficiently. The proposed method is an ensemble solution to sentiment analysis in textual data combining LSTM networks with a BERT-embedded layer. The model can capture the contextual nuances and sequential patterns of language. This paper employed two different social sentiment analysis textual datasets of depression and self-harm respectively. The proposed method gives an accuracy of $\mathbf{9 8. 4 \%}$ for the depression dataset and $\mathbf{9 2. 3 \%}$ for the self-harm dataset. Those datasets also used other machine learning approaches such as LSTM with a convolutional layer, BERT, Logistic Regression, and combined LSTM and BERT. The proposed method significantly improves accuracy compared to existing state-of-the-art methods and models. To ensure immediate action and quick recovery and minimize possible serious instances like suicide, this research attempts to identify early indicators of depression.