STUDENTLERDI BIRLESTIRIW: XALÍQARALÍQ IZERTLEWLER HÁM PÁNLER BOYINSHA BIRGE
ISLESIW 1-XALÍQARALÍQ STUDENTLER KONFERENCIYASÍ. NÓKIS, 2025-JÍL 20-21-MAY
__
_____________________________________________________________________________________________
428
AI FOR PREDICTING STUDENT MENTAL HEALTH CRISES: A DATA-DRIVEN
APPROACH FOR EARLY INTERVENTION
UMOH, OTO-OBONG EZEKIEL
Msc-Information Systems, Assistant Lecturer,
Yessenov University, Aktau, Kazakhstan.
INTRODUCTION.
The rise of mental health crises among students is gradually becoming a
menace globally because it does not just affect academic performance but also deep-seated social and
personal implications [1;2;3], hence the necessity for innovative solutions. According to the World
Health Organization, approximately 20% of teenagers and young adults tend to experience mental
health challenges (with a greater portion left undiagnosed or untreated) [10] due to a couple of reasons
which includes academic pressure, financial concerns, social integration, and uncertainties as it
concerns future employment. Conventional evaluation techniques like counsellor-to-student and self-
disclosure; that depends on self-reported surveys and clinical evaluations, often fail to provide timely
interventions.
Artificial intelligence (AI) technologies offer a promising solution to revolutionize mental health
support by analysing diverse data sources, including health surveys, electronic records, and even
social media activity, to anticipate crises before they escalate [4;5;6]. This study presents an AI-based
predictive tool that is designed to specifically assess the risk of student mental health crises with the
help of integrated data sources which includes behavioural, academic, and lifestyle indicators.
As this field continues to grow, there are ongoing research, and interdisciplinary collaborations
that will be essential to refine predictive models and ensure the ethical deployment of AI technologies
in mental health contexts. In the future, the research will focus on navigating regulatory challenges,
enhancing international cooperation, and developing comprehensive training for educators and mental
health professionals to maximize the benefits of AI in supporting student well-being [7;8;9].
METHODOLOGY
Figure 1
. Research Procedure and Model Architecture
The study used different steps beginning with data collection from Kaggle with the title
“Student
Mental Health”
which includes responses from over 100 university students on topics including
academic stress, sleep patterns, and emotional well-being. The dataset comprises both numerical and
textual responses on students’ gender, academic performance, course of study, and mental health [11].
Fig.1 shows the different steps followed in this research. The dataset includes the following variables:
•
Demographics:
Gender, Age, Marital Status
•
Academic Profile:
Course, Year of Study, CGPA
•
Mental Health Indicatorss:
Self-reported Depression, Anxiety, Paniv Attacks
STUDENTLERDI BIRLESTIRIW: XALÍQARALÍQ IZERTLEWLER HÁM PÁNLER BOYINSHA BIRGE
ISLESIW 1-XALÍQARALÍQ STUDENTLER KONFERENCIYASÍ. NÓKIS, 2025-JÍL 20-21-MAY
__
_____________________________________________________________________________________________
429
•
Utilization of Support Services:
Specialist Treatment Sought (Yes/No)
'Age' by 'Gender' for 'Gender' 'Male' or 'Female'
Gender
Count of Age
Female
75
Male
25
Grand Total
100
0
10
20
30
40
50
60
70
80
Female
Male
Age
G
e
n
d
e
r
'Age' by 'Gender' for 'Gender' 'Male' or 'Female'
Age
18
19
20
21
22
23
24
Figure 2
. Age distribution of the students by Gender
Preprocessing (Data Cleaning)
After the data collection, the next step was
data cleaning
. Fig. 2 shows the distribution of gender
by age after the data was cleaned using Pandas, with missing values removed.
Data Exploration
•
Prevalence:
Depression (≈30%), Anxiety (≈35%), Panic Attacks (≈20%).
•
Help-Seeking:
<10% of symptomatic students accessed specialist care.
•
Academic Correlates:
Lower CGPA (<3.0) is modestly associated with increased mental
health symptoms.
•
Demographic Patterns:
Female students and those in transitional academic years (first and
final year) exhibit higher rates of distress.
AI METHODOLOGIES FOR PREDICTING MENTAL HEALTH RISK
Model Selection
•
Logistic Regression:
This is most suitable for binary classification because of its high level
of interpretability. A
recall
of
1.00
for
class 0 (absence of mental health condition),
On the other
hand, the
recall
on
class 1
was low at
0.12
(
presence of mental health condition),
while the F1-Score
for
class1
was
0.22
and an overall accuracy of
0.67.
•
Random Forests:
This combines many decision trees to improve predictive accuracy and
reduce the risk of overfitting. An
accuracy
of
71%,
with a
recall
of
1.00
for
class 0
and
0.25
for
class 1,
and an
F1-Score
of
0.40
for
class 1
; indicating that it fails to identify a significant number
of actual positives.
•
Gradient Boosting:
It combines multiple decision trees in a sequence, with each tree
correcting errors made by its predecessor. It shows an
F1-Score
of
0.50
for
class 1
, with the overall
accuracy
of
0.71
. The recall for class 0 and class 1 were
0.92
and
0.38
respectively.
•
Neural Networks:
This is a very powerful model for large and complex data, but because of
the limited amount of dataset used in this research, it achieved a of
0.00
for
recall
and
F1-Score
in
class 1
.
STUDENTLERDI BIRLESTIRIW: XALÍQARALÍQ IZERTLEWLER HÁM PÁNLER BOYINSHA BIRGE
ISLESIW 1-XALÍQARALÍQ STUDENTLER KONFERENCIYASÍ. NÓKIS, 2025-JÍL 20-21-MAY
__
_____________________________________________________________________________________________
430
Table 1.
Model Performance
Feature Engineering
This process involves selecting, modifying, and creating new features from raw data, hence
improving the ability of the model to capture patterns. From the above model(s) selection, we saw
that there were struggles particularly in
class 1
-
indicating the presence of mental health issues.
Key Features and Their Importance
Academic Performance Metrics:
•
CGPA/Year of Study:
These quantitative indicators of students can correlate with stress
and mental health. It is transformed from a range (e.g., “3.00 - 3.49”) into a numerical average,
where a
High CGPA
could mean
lower mental health issues.
Demographic Information:
•
Age, Gender, Marital Status:
Categorizing the ages can give deeper insights on the mental
state of students e.g. 18-22, 23-25. Likewise, we encoded gender as a numerical value(s) [0 for
Female, 1for Male] and their marital status.
Mental Health Indicators:
•
Depression, Anxiety, Panic-Attack:
These are binary indicators of students’ mental status
which are converted to numerical values (1 for Yes, 0 for No); these can be analysed using
demographic and academic variables.
Implementation of Feature Engineering
To enhance the model selection using feature engineering, we wrote down some codes in
python after installing the necessary libraries like pandas (to access the dataset-.csv), matplotlib
(visualization) and seaborn (conversion).
Model
Precisio
n (0)
Recal
l (0)
F1-
Scor
e (0)
Suppor
t (0)
Precisio
n (1)
Recal
l (1)
F1-
Scor
e (1)
Suppor
t (1)
Accurac
y
Macr
o
Avg.
F1
Weighte
d Avg.
F1
Logistic
Regressio
n (LR)
0.65
1.00
0.79
13
1.00
0.12
0.22
8
0.67
0.51
0.57
Random
Forest
(RF)
0.68
1.00
0.81
13
1.00
0.25
0.40
8
0.71
0.61
0.66
Gradient
Boosting
(GB)
0.71
0.92
0.80
13
0.75
0.38
0.50
8
0.71
0.65
0.69
Neural
Network
(NN)
0.62
1.00
0.76
13
0.00
0.00
0.08
8
0.62
0.38
0.47
STUDENTLERDI BIRLESTIRIW: XALÍQARALÍQ IZERTLEWLER HÁM PÁNLER BOYINSHA BIRGE
ISLESIW 1-XALÍQARALÍQ STUDENTLER KONFERENCIYASÍ. NÓKIS, 2025-JÍL 20-21-MAY
__
_____________________________________________________________________________________________
431
Figure 3.
Comparing CGPA and Year of Study
Insights From the Datasets
Class Imbalance and Model Performance:
After the analysing the dataset, it was observed
that the models show varying performance between
class 0
and
class 1.
To illustrate, whereas all
models got higher
recall
for
class 0
(non-mental health issues), they still struggled with
class 1
(mental health issues) especially in the low
precision
and
recall
for
class 1
in models like Neural
Network scoring 0.00 for both metrics.
Best Performance:
Gradient Boosting from the model demonstrates to be the best balance with
an
F1-score
of 80% and
class 0
of 50% for
class 1
. This implies that it has the capacity to capture
more complexities in data in comparison to others.
Impact of Academic Performance:
As seen in the chart above (Fig.3), the presence of mental
health issues among students is closely related to the different stages or levels of learning due to more
academic pressures.
Need for Targeted Interventions:
Having seen how demographic factors and academic
performance can greatly influence mental health, it is advisable that universities should consider
implementing targeted mental health initiatives focusing on specific courses, age groups and year of
study.
DISCUSSION
Summary of Key Findings:
In this research, Gradient Boosting was the best performing model
with an
F1-score
of 80% for
class 0
and 50%
class 1
. This means that it can balance precision and
recall effectively. It also suggests the importance of academic achievement as a possible predictor of
mental health status. This finding corresponds with existing literature that emphasizes academic stress
as a significant factor in mental health issues among the student population [12;13].
Mental Health Policy and Practice Implications
The ability to identify early warning signs of mental health issues through predictive modelling
can facilitate early intervention, which can potentially minimize the severity of crises. Schools could
use such knowledge to develop tailored mental health support services for specific disciplines or
courses of study. For instance, students with low GPAs might have to be provided with additional
counselling and academic support, wherein their academic performance and psychological well-being
are addressed simultaneously. By identifying at-risk groups, educational authorities can cautiously
distribute mental health resources better, with support services being delivered where they are most
required [14].
Ethical Considerations
and Limitations
STUDENTLERDI BIRLESTIRIW: XALÍQARALÍQ IZERTLEWLER HÁM PÁNLER BOYINSHA BIRGE
ISLESIW 1-XALÍQARALÍQ STUDENTLER KONFERENCIYASÍ. NÓKIS, 2025-JÍL 20-21-MAY
__
_____________________________________________________________________________________________
432
Imbalance in Class Distribution:
The large imbalance in class distribution found in the dataset
implies that future studies must be aimed at ways to improve model performance on the minority class
possibly using SMOTE [15].
Privacy Issues:
Institutions need to ensure that they are following ethical standards and data
protection regulations [16].
Human Oversight:
The use of artificial intelligence to forecast mental health outcomes
underscores the need for human intervention in the decision-making process. Clinical judgment
should always have a primary role in assessing the suitability of any intervention based on AI
predictions.
Real-World Implementations and Strategies for Intervention
Early Warning Systems:
Educational institutions can develop artificial intelligence-powered
early warning systems that alert mental health professionals when students exhibit signs of distress
for effective clinical responses.
Integration with Support Services:
Artificial intelligence models can be integrated into existing
student support systems, enabling the early identification of students who may need additional
resources.
Targeted Awareness Initiatives:
Data derived from predictive models can guide the creation of
tailored mental health awareness initiatives for specific demographic populations or academic schools.
Opt-in Monitoring Programs:
Institutions can provide opt-in monitoring for students who have
a history of mental health problems with users’ consent.
Improved Clinical Evaluation:
Mental health practitioners may utilize artificial intelligence
results to improve clinical assessments, thereby gaining a better insight into student well-being.
CONCLUSION:
The study demonstrates the promise of machine learning models in identifying
problems with students' mental health. Our research, considering demographic factors, leads also to
important variables influencing academic success; this makes it possible to envision quite an early
and effective intervention in mental health. The ethical concerns and challenges raised in the present
work, however, make evident the need for scrupulous implementation and ongoing research.
Balancing artificial intelligence's prediction potential and student confidentiality individual cases
would be essential in crisis and mental health intervention.
REFERENCES:
1.
World Health Organization. Global health estimates: mental disorders. 2024.
2.
Garriga, R., et al. Machine learning model to predict mental health crises from electronic
health records. Nat Med. 2022.
3.
Dachew, B.A., et al. Prevalence of mental distress among university students. J Ment Health.
2022.
4.
Evaluating predictive artificial intelligence approaches used in ...
5.
Recognizing the Signs of Mental Health Crisis in Students
6.
Study: AI tool can help counselors predict which college students ...
7.
When mental health meets technology: Faculty and students ...
8.
Artificial Intelligence, Data Privacy, and How to Keep Patients Safe ...
9.
AI in Predicting Mental Health Trends - LinkedIn
10.
World Health Organization. (2021). Adolescent mental health.
11.
12.
Beiter, R., Nash, R., McCrady, M., et al. (2015). The impact of stress on college students.
Journal of American College Health
, 63(2), 80-86.
13.
Eisenberg, D., Golberstein, E., & Gollust, S. E. (2009). The impact of mental health stigma
on seeking and participating in mental health care.
Health Affairs
, 28(3), 882-887.
STUDENTLERDI BIRLESTIRIW: XALÍQARALÍQ IZERTLEWLER HÁM PÁNLER BOYINSHA BIRGE
ISLESIW 1-XALÍQARALÍQ STUDENTLER KONFERENCIYASÍ. NÓKIS, 2025-JÍL 20-21-MAY
__
_____________________________________________________________________________________________
433
14.
Gonzalez, A., et al. (2020). The importance of mental health resource allocation in
educational settings.
Public Health Reports
, 135(1), 14-22.
15.
Chawla, N. V., De Silva, V., & Van der Laan, M. (2002). Adapting classification algorithms
to imbalanced data sets.
Proceedings of the 2002 International Conference on Knowledge
Discovery and Data Mining
, 244-249.
16.
Regulation (EU) 2016/679. (2016). General Data Protection Regulation (GDPR).
THE USE OF ESPERANTO IN THE DEVELOPMENT OF METALANGUAGE
AWARENESS
Urazniyazova G.G,
The 3rd year student of
the Foreign languages faculty of KSU
Summary:
The article examines the features of the grammatical structure and syntax of
Esperanto, its importance in psycholinguistics, as well as its role in the theory of universal language.
The article considers the use of Esperanto in educational where it promotes the development of
metalanguage skills and facilitates the development of other languages.
Key words:
linguistic theories, assimilation, era of globalization, medium, educational
interaction.
Esperanto is an artificial language created at the end of the 19th century by the Polish linguist
Ludwik Zamenhof. Despite the fact that this language did not become a global means of
communication, as originally planned, it had a significant impact on the development of linguistics,
as well as on world linguistic culture. In this article, we will look at how Esperanto influences language
learning, the development of linguistic theories and techniques, as well as its cultural and educational
role [1].
Initially, Esperanto was conceived as a universal language that could serve as an international
means of communication, promoting peace and mutual understanding between peoples.
However, even if this ideal has not been fully achieved, Esperanto has become an important object
for linguistic research. One of the key features of Esperanto is its constructivism: the language was
designed with maximum logic and simplicity in mind. Grammatical structure and syntax in Esperanto
offers unique opportunities for exploring language structures. Its grammar contains no exceptions,
which makes its study less difficult compared to natural languages. Esperanto has a fairly simple
morphology, as well as a unified system of declensions and conjugations. This provides linguists with
a convenient tool to study the interaction of different language elements such as roots, prefixes, and
suffixes, as well as their role in the formation of new words [6].
Some researches in the field of psycholinguistics also actively uses Esperanto to understand how
people perceive and assimilate language structures. Esperanto can serve as an ideal material for
experiments, as its simplicity and regularity allow us to explore cognitive processes such as the
perception of grammatical errors or the speed of language acquisition. Esperanto in the context of the
theory of universal language Esperanto continues to play an important role in discussions about the
possibility of creating a universal language.[2].
Linguists such as Noam Chomsky and his followers have considered hypotheses about the
existence of a universal grammar that could underlie all human languages. Esperanto, with its regular
grammar and universal elements, serves as an interesting example of an artificially created language
that could emdiv the principles of universality. Esperanto also provides an opportunity to explore
the concepts of "linguistic universalism" and "linguistic relativity." Languages designed to be neutral
