Table 5 Analyzing the efficacy of machine-learning models in detecting depression through social media data.

Yazdavar et al.⁷⁴

Achieved 68% accuracy and 72% precision in identifying clinical depressive symptoms using a semi-supervised statistical model.

Zogan et al.⁷⁵

Proposed a new computational model and achieved a recall of 0.904, precision of 0.909, and F1 score of 0.912.

Aragón et al.⁵⁷

Using fine-grained emotions to obtain competitive results in comparison to state-of-the-art approaches.

Paul et al.³²

AdaBoost classifier outperformed other methods for depression likelihood assessment.

Ricard et al.¹⁶⁹

Leveraging community-generated content from social media can be informative for automated depression assessment.

Peng et al.¹⁷⁰

Demonstrated that a multi-kernel support vector machine is the most appropriate approach to identifying depression in individuals using social media.

Aldarwish et al.¹⁷¹

Trained a support vector machine based on term frequency to classify depression levels.

Chiong et al.³¹

The proposed model effectively determines depression presence via social media posts, even when the training datasets do not contain depression-related words.

Burdisso et al.¹⁷²

Introduced a general framework for early depression detection with less computational cost and higher interpretability.

Smys et al.¹⁷³

A machine-learning model consisting of a support vector machine and a naive Bayes model can predict depression in its early stages.

Bucur et al.¹⁷⁴

Latent semantic analysis shows a significant difference in writing topics depending on users’ mental health.

Kayalvizhi et al.¹⁷⁵

A word2vec pre-trained word embedding and random forest classifier achieved their best performance with a 0.877 F₁ score.

Mann et al.¹⁷⁶

Fusion model can detect moderate depression or higher with 0.92 recall and 0.69 precision.

Sadeque et al.¹⁷⁷

Proposed a system to effectively detect depression using social media content with an accuracy of 88% and F₁ score of 93%.

Hussain et al.¹⁵²

Application accurately identifies indicators of depression in Facebook users with 94% accuracy.

Tadesse et al.⁵⁸

Achieved 91% accuracy and F₁ score of 93% with a multi-layer perceptron algorithm and combined features.

Fatima et al.¹⁷⁸

Achieved an accuracy, recall, and precision of 91.7% using a combination of text-based features and machine-learning techniques.

Katchapakirin et al.¹⁷⁹

Facebook behaviors can be used to predict depression levels with an accuracy of 85% and F₁ score of 88.9%.

Shen et al.¹⁸⁰

The model outperformed several baselines by 3% to 10% with an F₁ score of 85%.

Li et al.¹⁸¹

Proposed a correlation explanation learning algorithm to detect COVID-19-related stress symptoms.

Lin et al.¹⁸²

Social media use is significantly associated with increased depression risk.