Predicating depression on Twitter using hybrid model BiLSTM-XGBOOST

Abstract

Nowadays, depression is a common mental illness. Failure to recognize depression early or guarantee that a depressed individual receives prompt counseling can lead to serious issues. Social media allow us to monitor people's thoughts, daily activities, and emotions, including persons with mental illnesses. This study suggested novel hybrid models that combine one of the deep learning techniques with one of the machine learning approaches. This paper used a dataset from the Kaggle website to predict depression. Two deep learning techniques were chosen to conduct the experiments: bidirectional long short-term memory (Bi-LSTM), and convolutional neural network (CNN). Three machine learning techniques were also selected, which are support vector machine (SVM), light gradient boosting machine (LGBM), and extreme gradient boosting (XGBOOST). Deep learning methods were applied to extract important features from input data and training, and then machine learning was utilized to predict the class. The performance of the hybrid models was compared against that of five single models. The results showed that Bi-LSTM-XGBOOST is better than single models and achieve the highest performance, with 94% for all evaluation metrics. The proposed model can improve the performance of machine learning techniques and increase the detection rate of depression.