Table 1. A comparison between state-of-the-art depression detection techniques.

Paper	Year	Technique	Language	Data source	Data size	Accuracy (%)
Victor et al. (2019)	2019	BiLSTM	English	Questionnaire	671 participants	70.88
Rosa et al. (2018)	2019	CNN+BiLSTM	Portuguese	Questionnaire	146 participants	90
Wang, Niu & Yu (2019)	2020	Sentiment diffusion patterns	English	Twitter	100,000 tweets	79
Akhtar et al. (2019)	2020	CNN+LSTM+GRU	English	EmoInt	7,102	89.88
				EmoBank	10,062
				Facebook post	2,895
				SemEval-2016	28,632
Suman et al. (2020)	2020	DL	English	Twitter	1,600,000	87.23
Shetty et al. (2020)	2020	LSTM	English	Twitter	N/A	70
		Model vector				76.69
Alabdulkreem (2021)	2021	RNN-LSTM	Arabic	Twitter	10,000	72
Chiu et al. (2021)	2021	CNN for image	English	Instagram	520 users	N/A
		Word2vec for text
		Handcrafted features for behavior
Tommasel et al. (2021)	2021	RNN	Spanish	Twitter	150 million tweets	N/A
Jyothi Prasanth, Dhalia Sweetlin & Sruthi (2022)	2022	RNN	English	Twitter	1,200 users	72
		LSTM				76
		BiLSTM				90
Zogan et al. (2022)	2022	CNN+RNN	English	Twitter	4,208 users	89.5
Kour & Gupta (2022)	2022	CNN-biLSTM	English	Twitter	1,402 depressed	94.28
					300 million non-depressed
Nair et al. (2022)	2022	Machine learning	English	Twitter	10,000 tweets	97
Park & Moon (2022)	2022	BERT-CNN	English	DAIC-WOZ	N/A	N/A
Safa, Bayat & Moghtader (2022)	2022	Machine learning	English	Twitter	11,890,632 tweets	91
Lia et al. (2022)	2022	Machine learning	English	Twitter	1,600,000	79.9
Shankdhar, Mishra & Shukla (2022)	2022	BiLSTM+CNN	English	Twitter	12,274 tweets	95.12
Tong et al. (2022)	2022	Discrete Adaboost Cost-sensitive Boosting Pruning Trees	English	TTDD	7,873 tweets	88.39
				CLPsych 2015	1,746 tweets	70.69
				LSVT	128	85.72
				Statlog	6,435	92.21
				Glass	214	77.63