Feature-based enhanced boosting algorithm for depression detection

Depression is a rapidly increasing mental disorder that can interfere with a person’s ability and negatively affect functions in various aspects of life. Fortunately, machine learning and deep learning techniques have demonstrated excellent results in the early detection of depression using social m...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:PeerJ. Computer science Ročník 11; s. e2981
Hlavní autoři: Rohei, Muhammad Sadiq, Varathan, Kasturi Dewi, Palaiahnakote, Shivakumara, Anuar, Nor Badrul
Médium: Journal Article
Jazyk:angličtina
Vydáno: United States PeerJ. Ltd 29.07.2025
PeerJ Inc
Témata:
Algorithms
Analysis
Depression detection
Depression, Mental
Enhanced boosting algorithm
Feature engineering
Feature-based enhanced boosting algorithm
Machine learning
Social media
Taiwan
Depression detection
Enhanced boosting algorithm
Feature-based enhanced boosting algorithm
Feature engineering
ISSN:2376-5992, 2376-5992
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:Depression is a rapidly increasing mental disorder that can interfere with a person’s ability and negatively affect functions in various aspects of life. Fortunately, machine learning and deep learning techniques have demonstrated excellent results in the early detection of depression using social media data. Most recently, researchers have utilized boosting algorithms including pre-defined boosting algorithms or built their own boosting algorithm for the detection of depression. However, both types of boosting algorithms struggle with the analysis of complex feature sets, the enhancement of weak learners, and the handling of larger datasets. Thus, this study has developed a novel feature-based enhanced boosting algorithm (F-EBA). The proposed model covers two pipelines, the feature engineering pipeline which improves the quality of features by picking up the most relevant features while the classification pipeline uses an ensemble approach designed to boost/elevate the model’s performances. The experimental results highlighted that various parameter including WordVec and BERT embeddings, attention mechanisms, and feature elimination techniques, significantly contributed to the selection of the most relevant features. This approach resulted in generating an optimized feature set that augmented both the model’s accuracy and its interpretability. In addition, utilizing over 46 million records, the F-EBA model significantly enhanced the performance of weak learners through a weight maximization strategy, achieving an impressive accuracy rate of 95%. Moreover, the integration of an adversarial layer that employs defense mechanisms against synonymous text and sarcastic phrases within the datasets has further boosted the F-EBA model’s accuracy to approximately 97%, surpassing the results reported in prior studies. Moreover, the optimized feature sets derived from the F-EBA model make a substantial contribution to boosting the performance of baseline classifiers, marking a novel advancement in the field.
Bibliografie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 23
ISSN:2376-5992
2376-5992
DOI:10.7717/peerj-cs.2981