A Hybrid Stacked Ensemble Technique to Improve Classification Accuracy for Neurological Disorder Detection on Reddit posts

Abstract

Sentiment analysis helps in the early detection of depression as identify unpleasant mental states in people who are at risk for developing mental disorders. By utilizing both syntactic and semantic information, modelling approaches for sentiment analysis rely on machine learning algorithms. In this paper, a hybrid stacked ensemble learning approach has been used for the detection of depression as a neurological disorder. With the help of pre-trained word embeddings, the Word2Vec, GloVe and Fasttext are chosen for data preprocessing and feature extraction. Then, to identify depressed and non depressed identities, we integrate a hybrid stacked ensemble learning approach over Random Forest (RF), Support vector machines (SVM), K-Nearest Neighbor (KNN) and Catboost classifier (CBC) as base models whereas logistic regression (LR) as meta model classifier. The results of the experiments show that suggested model performs best with our proposed model than individual models. It is also found that with Word2Vec word embedding model, the proposed model achieved the higher accuracy as 99% in comparison to GloVe and Fasttext that categorizes depressed over non depressed users on the social media platforms.