Integrating Deep Learning into Educational Wellbeing: Early Screening of Anxiety, Depression, and Stress Among University Students

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This article deals with a topic of current interest, which is important in the field of education in particular and for society in general. The design, methodology and results are very comprehensive and well described. Similarly, the references used to construct the theoretical framework and discussion are current and relevant. 
However, on the other hand, the objectives are not described, nor are there any research questions or hypotheses. This makes it very difficult to determine the purpose of the research. I believe that this part is essential for a complete research study. Therefore, it is necessary to add it. 

Author Response

Please see the attached file.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The article is worthy of publication but only after some minor changes.

The authors have performed some experiments to show that AI and ANN can be of help with mental problems of students. They argue they have achieved 95% of accuracy when compared to other approaches. This make s the article important and interesting.

The minor changes that are needed are as detailed hereunder.

Figure 1 could preferably be on the same page as the fragment that is relevant to it , unless there  is a good reason not to position it there.

Some language improvement and proofing will be helpful. For instance: "....as show in Figure 1..." on line 70.

In Figure 1 some labels are in Spanish. It would be better to translate all into English as the article is in English. The authors can't presume that all reader lnow that IA (inteligencia artificial) means AI and neuronal means neural.

Figure 1 not really described, explained, lacks legend. 

The Discussion and Conclusion sections are too short on authors' inferences, ideas, approaches and suggestions. The authors would make the article much better if they could share with the readers their own knowledge on the subject.

The article length allows for much more elaboration on authors' views and suggestions (the article''s length is 5503 words of which technical metadata and  references are 1672, leaving the substantive text at only 3831 words).

The first part of Discussion would be better positioned in Literature Review or State of the art sections, as it describes those rather than discussion of authors innovative personal views:

"This study is framed within a growing global trend in which artificial intelligence 186
(AI), particularly deep neural networks and machine learning models, has emerged as a 187
valuable tool for the diagnosis, monitoring, and prevention of common mental disorders 188
such as anxiety, depression, and stress in university populations [19 –21 ]. This responds to 189
the sustained rise in prevalence, the limited availability of specialized care—especially in 190
low- and middle-income countries—and the increasing demand for scalable, personalized 191
interventions. 192
Recent literature emphasizes the importance of personalization. Systematic reviews 193
and empirical studies show that new AI systems not only automate diagnosis but also 194
identify individual risk factors and design preventive strategies tailored to vulnerable 195
groups such as university students, who often face high academic and social pressures 196
[22 ]. In this context, adaptive technologies, including chatbots, are reshaping mental 197
health care by supporting both prevention and remote follow-up [23]. Recent AI-based 198
models have successfully generated recommendations based on demographic, lifestyle, 199
Version October 18, 2025 submitted to Journal Not Specified 9 of 12..
Figure 10. Results of neural network training with real data.
and psychometric data, reaching high accuracy in predicting and preventing psychological 200
distress [24]. 201
Despite this potential, the implementation of AI in mental health entails ethical and 202
technical challenges. Issues such as algorithmic bias, limited generalizability, and defi- 203
ciencies in validation and reporting remain significant concerns [12 ,25 ,26 ]. International 204
guidelines recommend strict data protection, robust governance, human oversight, and 205
clear definition of application contexts [ 26 – 28]. Multidisciplinary collaboration, socio- 206
cultural adaptation, and fairness audits are also necessary to ensure responsible adoption 207
[29 , 30]. While many studies report accuracies above 80% for disorders such as depres- 208
sion and suicide risk, the need to expand population diversity and strengthen model 209
interpretability remains urgent [15,31]. 210
From a technical standpoint, multimodal models combining voice, text, behavioral, 211
and neuropsychological data can detect early signs of disorders even before clinical man- 212
ifestation [19 , 32]. Performance consistently depends on the choice of architecture and 213
parameters such as number of layers, activation functions, and training epochs [33, 34 ]. 214
Moderately deep, well-regularized networks (e.g., ResNet) with ReLU activations optimize 215
training and mitigate saturation issues [35– 37 ], while random search often proves more 216
efficient than grid search for hyperparameter optimization [38]. 217"

Comments on the Quality of English Language

Some language improvement and proofing will be helpful. For instance: "....as show in Figure 1..." on line 70.

In Figure 1 some labels are in Spanish. It would be better to translate all into English as the article is in English. The authors can't presume that all reader lnow that IA (inteligencia artificial) means AI and neuronal means neural.

Author Response

Please see the attached file.

Author Response File: Author Response.pdf