1. Introduction
Young college students often struggle to cope with negative emotions because of the
abrupt changes in their environment and the increasing pressures from various sources
[1]. Issues such as anxiety and depression have become increasingly prevalent among this
group [2], alongside common problems like eating disorders and sleep disturbances [3]. Untreated mental health issues can escalate to self-harm, suicide, and other irreversible
consequences [4]. With the advances in big data technology, making accurate predictions regarding
the mental health of college students would be beneficial for schools to monitor changes
in their well-being and strengthen student management. Currently, there are several
methods to predict anxiety, depression, and other mood disorders [5]. Tan et al. [6] introduced an objective data annotation method and developed a long short-term memory
(LSTM) model to detect anxiety states that outperform traditional algorithms. Akbari
et al. [7] extracted features from electroencephalogram signals and employed a support vector
machine and K-nearest neighbor (KNN) classifier to distinguish normal and depressed
signals. They reported that the classifier achieved an accuracy of 98.79% and could
be utilized for early depression diagnosis. Sardari et al. [8] combined a convolutional neural network and self-encoder to learn the features from
audio sources for depression detection, increasing the F value by at least 7%. Fauziah
et al. [9] used the random forest and XGBoost to extract features from the YouTube comment text
for anxiety recognition, with both methods achieving 83% and 73% accuracy, respectively.
Based on big data technology, this paper provides a fresh perspective on the mental
health status of college students. This study collected data through a questionnaire
survey and designed an improved seagull optimization algorithm-back-propagation neural
network (ISOA-BPNN) prediction model to predict the mental health of college students.
Experimental analysis demonstrated the effectiveness of this method for predicting
mental health, providing a reliable approach for managing student emotions in universities.
This method is beneficial for better understanding the changes in the mental health
conditions of college students, enabling early detection, intervention, and assistance
for students who may face psychological issues. The method offers scientific decision-making
support for learning and helps schools improve the effectiveness of their mental health
services. Ultimately, it holds significant social value.
2. Big Data Analysis of College Students’ Mental Health
2.1 Factors Affecting the Mental Health of University Students
As integral members of society, college students face pressures from diverse sources,
including society, academia, and family. Being highly educated individuals, college
students are subject to elevated external expectations and impose higher standards
on themselves, resulting in more frequent experiences of negative emotions and an
increased likelihood of encountering mental health issues. Based on real-life interviews
and a comprehensive literature review, this paper identified pressure sources as the
central factors influencing the mental well-being of college students and designed
a questionnaire involving their psychological pressure sources.
A questionnaire was developed based on the above ten questions. The subjects quantified
the impact of these ten pressure sources on their mental health status according to
their situation, and a five-point Likert scale was used.
2.2 SCL-90
The Symptom Checklist 90 (SCL-90) is a widely employed instrument for assessing mental
health status [10]. It boasts well-established reliability and validity. It can be applied to different
age groups and cultural populations, demonstrating good generalizability and flexibility.
The SCL-90 covers the assessment of multiple domains of psychological symptoms and
possesses a certain level of objectivity. Therefore, it was used to evaluate the mental
health status of college students. This scale comprised 90 items, each rated on a
five-point scale. The score ranging from 1 to 5 represents the impact of the item
on oneself (from small to large). This rating scale is illustrated in Table 2.
Table 1. Analysis of Influencing Factors of College Students’ Mental Health.
|
Pressure source classification
|
Pressure source division
|
Questionnaire format
|
|
Employment pressure source
|
X1: Employment pressure from society
|
Under intensified social competition, the imbalance between the supply and demand
of talents, and the declining employment advantage of college students, what pressure
do you feel?
|
|
X2: Employment pressure from school
|
Under the school's course arrangement on employment policy analysis and career planning,
what pressure do you feel?
|
|
X3: Employment pressure from self
|
Among individual employment skills, such as theoretical knowledge, research abilities,
and innovative spirit, what pressure do you feel?
|
|
Academic pressure source
|
X4: Academic pressure from school
|
In the course of your study, do you feel that the school's training objectives and
curriculum are unreasonable and the teaching methods are too single? What pressure
do you feel?
|
|
X5: Academic pressure from self
|
In the process of study, do you feel overburdened? Do you feel overwhelmed by the
demands of certificate examination, postgraduate examination, level examination, and
thesis writing? What pressure do you feel?
|
|
Economic pressure source
|
X6: Economic pressure from school
|
What economic pressures do the school's tuition and other expenses bring?
|
|
X7: Economic pressure from self
|
During the school year, faced with spending due to study, living, entertainment, etc.,
what pressures do you feel?
|
|
Interpersonal pressure source
|
X8: Pressure from teacher-student relationship
|
What pressure do you experience when communicating and interacting with teachers?
|
|
X9: Pressure from classmate relationship
|
What pressure does interacting with classmates bring to you?
|
|
X10: Pressure from love relationships
|
What do you feel when faced with romantic relationships, mate selection, and marriage?
|
Table 2. Correspondence between Factors and Question Numbers in SCL-90 [11].
|
Factor
|
Corresponding question number
|
|
Somatization
|
41, 4, 12, 27, 40, 42, 48, 49, 52, 53, 56, 58
|
|
Anxiety
|
2, 17, 23, 33, 39, 57, 72, 78, 80, 86
|
|
Mental disorder
|
7, 16, 35, 62, 77, 84, 85, 87, 88, 90
|
|
Interpersonal sensitivity
|
6, 21, 34, 36, 67, 41, 61, 69, 73
|
|
Hostility
|
11, 24, 63, 67, 74, 81
|
|
Fear
|
13, 25, 47, 50, 70, 75, 82
|
|
Compulsion
|
3, 9, 10, 28, 35, 45, 46, 51, 55, 65
|
|
Depression
|
5, 14, 15, 20, 22, 62, 29, 30, 31, 32, 54, 71, 79
|
|
Paranoia
|
8, 8, 43, 68, 76, 83
|
2.3 Data Analysis
An online questionnaire was disseminated to students from universities in Hebei Province.
The participants were requested to complete the psychological pressure source questionnaire
and the SCL-90. In total, 21,354 valid questionnaires were collected, and Table 3 lists some of the preliminary results.
Table 3. Sample Data.
|
Sample number
|
Results of the psychological pressure source survey
|
Total score
|
|
X1
|
X2
|
X3
|
X4
|
X5
|
X6
|
X7
|
X8
|
X9
|
X10
|
|
1
|
1
|
2
|
1
|
2
|
1
|
1
|
1
|
1
|
1
|
1
|
120
|
|
2
|
2
|
2
|
3
|
3
|
4
|
2
|
3
|
4
|
2
|
2
|
156
|
|
3
|
1
|
1
|
1
|
1
|
1
|
1
|
1
|
1
|
1
|
1
|
102
|
|
4
|
3
|
4
|
3
|
3
|
4
|
3
|
4
|
4
|
3
|
4
|
251
|
|
5
|
2
|
3
|
4
|
2
|
2
|
3
|
4
|
2
|
4
|
3
|
178
|
|
......
|
|
21354
|
1
|
1
|
1
|
2
|
1
|
1
|
1
|
2
|
1
|
2
|
138
|
The mental health of college students was analyzed based on the SCL-90 results. Fig. 1 presents the volume of data with an average score above two points for each factor.
Fig. 1. Volume of data with an average score above two points for each factor.
The most obvious psychological health problem among college students was compulsions
(Fig. 1), with 4,726 data, accounting for 22.13% of the total sample. Interpersonal sensitivity
was the second most common issue, with 3,561 data, accounting for 16.68% of the total
sample. Hence, college students have a high likelihood of experiencing symptoms of
compulsions and face difficulties in handling interpersonal relationships. Moreover,
depression and anxiety were also common among these college students, with proportions
of 10.65% and 10.46%, respectively.
The total score of the SCL-90 scale was then analyzed. In general, a score exceeding
160 points indicates the presence of certain psychological health issues, while a
score exceeding 200 points suggests the presence of significant psychological health
problems. A score surpassing 250 points signifies severe psychological health issues.
Fig. 2 presents the distribution of the total scores.
Fig. 2. Distribution of the total scores.
Of the 21,354 samples, 18,152 data, i.e., 85% of the total sample, exhibited SCL-90
scores below 160 points (Fig. 2). This group of college students is less likely to face mental health problems. Next,
2,031 data, accounting for 9.51% of the total sample, fell within the total score
range of 160 points to 200 points, indicating that these students may experience some
mental health challenges that could benefit from psychological counseling. There were
1,027 data, accounting for 4.81% of the total sample, with total scores ranging from
200 points to 250 points. This group of college students appeared to exhibit more
pronounced mental health problems, requiring the attention and intervention of both
the school and parents. In contrast, 144 data, comprising 0.67% of the total sample,
displayed total scores exceeding 250 points. These college students face serious mental
health issues, and the school and parents must intervene promptly.
Combining Figs. 1 and 2, the data with a total score exceeding 160 points or an average
score above two points on any factor were categorized as ``abnormal,'' while the remaining
data were labeled as ``normal.'' This classification yielded 4,937 abnormal mental
health data, accounting for 23.12% of the total sample, and 16,417 normal data, accounting
for 76.88%.
A significant imbalance existed in these samples, which could pose challenges for
subsequent prediction modeling. The synthetic minority over-sampling technique (SMOTE)
was used to enhance predictive accuracy [12] and rectify the imbalance. As a result, the dataset was transformed to include 6,234
abnormal mental health data and 6,234 normal mental health data, achieving a balanced
positive-to-negative sample ratio of 1:1.
3. Prediction Model based on BPNN
Building on the analysis in the preceding section, this paper used the scores associated
with the ten pressure sources outlined in Table 1 to predict the mental health status of college students. The predicted outcomes were
categorized as ``abnormal'' or ``normal.'' Given the intricate nonlinear relationships
between various pressure sources and mental health status, this study selected the
artificial neural network (ANN) method, demonstrating exceptional computational capabilities
and excelling in handling nonlinear problems [13].
The BPNN algorithm is used widely among ANN algorithms [14], which reduces errors through backward error propagation and has excellent self-learning
capability. Table 4 lists the parameters involved in the BPNN algorithm.
Table 4. BPNN Parameters and Definitions.
|
Parameter
|
Definition
|
|
$x_n$
|
An input to the input layer
|
|
$h_i$
|
An input to the hidden layer
|
|
$h_o$
|
An output of the hidden layer
|
|
$y_i$
|
An input to the output layer
|
|
$y_o$
|
An output of the output layer
|
|
$w_{ih}$
|
Connection weight of input and hidden layers
|
|
$w_{ho}$
|
Connection weight of the hidden layer and the output layer
|
|
$b_h$
|
A threshold of the hidden layer
|
|
$b_o$
|
A threshold of the output layer
|
|
$d_o$
|
Desired output
|
The prediction process of the BPNN algorithm can be described as follows:
(1) Network parameters are initialized.
(2) The output of the hidden layer is calculated:
(3) The output of the output layer is calculated:
(4) The error between the BPNN output and the desired output is calculated:
(5) The weights and thresholds are constantly updated to determine if the error satisfies
the specific requirements. If not, return to step (2); if yes, the training ends.
The seagull optimization algorithm (SOA) was used to improve the BPNN algorithm and
avoid the defect that the BPNN easily falls into a local optimum. The principle of
SOA is to find the optimal solution by studying the migratory patterns and attacking
the behaviors exhibited by seagulls [15]. When the seagull population migrates, to avoid collision with neighboring seagulls,
it first needs to find a new position that does not conflict with other seagulls,
which can be written as
where $t$ represents the iteration count; $C_{S}\left(t\right)$ is the new position
where the individual has no conflict with other gulls; $H_{S}\left(t\right)$ represents
the position before migration; $f_{c}$ is a factor used to adjust the position of
the gulls to avoid collision.
where $t_{max}$ is the maximum number of iterations.
After collision avoidance, each gull approaches the optimal gull, which can be expressed
as
where $M_{s}\left(t\right)$ is the direction of the optimal gull; $H_{sbest}\left(t\right)$
is the position of the optimal gull; $f_{b}$ is a search equilibrium factor to adjust
the flight distance of the gulls:
where $r_{d}$ is a random number in the range of [0,1] that obeys a uniform distribution.
The distance between the individual and the optimal seagull can be written as follows:
After finding a food source through migration, the gulls approach their prey and attack
using a spiral motion, which can be expressed as
where $\theta $ is a random number obeying a uniform distribution in the range [0,2${\pi}$],
and $r$ is the radius of the seagull spiral flight.
where $u$ and $v$ are spiral coefficients, generally taken as 1.
Finally, the seagull’s attack position can be expressed as
The seagull position is continuously updated until the maximum number of iterations
is reached. The optimal result is output.
To solve the problem that SOA is sensitive to the output population, this paper introduced
Logistics chaotic mapping to initialize the sink population:
The levy flight strategy [16] is then used to replace the spiral motion in the SOA and improve the algorithm’s
searching ability of the algorithm. $levy\left(\beta \right)$ the relative distance
produced by the seagulls after levy flight:
where $\Gamma $ is the Gamma function. The attack position of the seagull after the
levy flight update can be written as
The ISOA algorithm was used to optimize the BPNN parameters and obtain the ISOA-BPNN
algorithm, which was used to predict college students’ mental health. The specific
procedures are as follows.
(1) The BPNN structure was set, and the number of nodes at each layer was determined.
(2) The ISOA algorithm optimized the BPNN parameters to obtain the optimal initial
weights and thresholds.
(3) The BPNN algorithm was trained using optimized parameters.
(4) Prediction of the mental health of college students was implemented using the
trained BPNN algorithm.
4. Results and Analysis
The tests were conducted on a computer running the Windows 10 operating system, equipped
with an Intel(R) Core(TM)i5-8400CPU@2.80GHz and 8 GB of random access memory (RAM).
The algorithms were simulated on MATLAB2020a software. The population size of SOA
was 30, and the iteration count was 500. $f_{c}=2$. In BPNN, the number of input layer
nodes was 12, and the number of output layer nodes was 1. The determination of the
number of hidden layer nodes was based on an empirical formula:
where $n$ and $l$ are the number of nodes in the input and output layers, and $a$
is a constant in [1,10]. The value of $m$ was 4–13. The data were substituted into
the BPNN algorithm for training 100 times. The mean square error (MSE) was averaged,
as shown in Table 5.
Table 5. Number of Hidden Layer Nodes and Average MSE.
|
Number of nodes in the hidden layer
|
Average MSE
|
|
4
|
2.125
|
|
5
|
2.036
|
|
6
|
1.985
|
|
7
|
1.957
|
|
8
|
1.865
|
|
9
|
1.811
|
|
10
|
1.792
|
|
11
|
1.732
|
|
12
|
1.925
|
|
13
|
2.133
|
The average MSE was minimized at 1.732 when the number of nodes in the hidden layer
was 11 (Table 5). Therefore, the final BPNN structure was 12-11-1.
Table 6. Confusion Matrix.
|
|
Predicted value
|
|
0
|
1
|
|
Actual value
|
0
|
TP
|
FN
|
|
1
|
FP
|
TN
|
The prediction effect was evaluated based on the confusion matrix (Table 6), with 0 indicating normal mental health status and 1 indicating an abnormal mental
health status. If the total score of SCL-90 exceeds 160 or the average score of any
factor exceeds 2, there may be psychological problems, such as anxiety and depression,
which require further attention and examination. Table 6 lists the evaluation indicators.
(1) Accuracy: $Accuracy=\frac{TP+TN}{TP+FP+FN+TN}$
(2) Precision: $Precision=\frac{TP}{TP+FP}$
(3) Recall rate: $Recall=\frac{TP}{TP+FN}$
(4) F1 value: the balance between precision and recall rate:$F1=\frac{2\times Precision\times
Recall}{Precision+Recall}$
(5) AUC: the area under the receiver operator characteristic (ROC) curve, which is
approximately close to 1, indicating better predictive performance.
First, the search capability of the ISOA algorithm designed in this paper was verified
by testing it on five benchmark functions (Table 7). Moreover, it was compared with other optimization algorithms, including:
(1) particle swarm optimization (PSO) algorithm [17],
(2) artificial bee colony (ABC) algorithm [18],
(3) sparrow search algorithm (SSA) [19].
The results are listed in Table 7.
Table 7. Benchmark Functions.
|
|
Function name
|
Domain of definition
|
Theoretical optimal value
|
|
F1
|
Sphere
|
[−100,100]
|
0
|
|
F2
|
Schwefel'problem 2.22
|
[−10,10]
|
0
|
|
F3
|
Schwefel'problem 2.21
|
[−100,100]
|
0
|
|
F4
|
Generalized Rosenbrock's Function
|
[−30,30]
|
0
|
|
F5
|
Step Function
|
[−100,100]
|
0
|
The size of the population of all the algorithms was set as 30, and the number of
iterations was set as 500. They ran 30 times, with the average taken. Table 8 lists the results.
Table 8. Comparative Results of the Benchmark Functions.
|
|
F1
|
F2
|
F3
|
F4
|
F5
|
|
PSO
|
9.74E+00
|
2.64E+01
|
7.43E+01
|
2.98E+06
|
6.50E03
|
|
ABC
|
6.81E-18
|
6.00E-12
|
6.37E-02
|
2.89E+01
|
4.21E-02
|
|
SSA
|
6.73E-01
|
2.83E-05
|
2.55E-01
|
2.87E+01
|
3.98E+00
|
|
SOA
|
2.26E-01
|
2.43E+01
|
1.76E+01
|
2.45E+04
|
3.94E-01
|
|
ISOA
|
5.67E-30
|
1.03E-17
|
1.81E-07
|
2.56E+01
|
2.12E-03
|
The bold values in Table 8 represent the optimal values. The ISOA was more capable of achieving the optimal
solution among the five test functions, suggesting that the ISOA had a stronger optimization
ability and better precision in problem-solving than the PSO, ABC, and other methods.
The impact of the ISOA on improving the BPNN prediction performance was evaluated,
as shown in Table 9.
Table 9. Improvement of the BPNN Algorithm Accuracy using Different Optimization Algorithms.
|
|
Accuracy
|
|
BPNN
|
0.8434
|
|
PSO-BPNN
|
0.8872 (+0.0438)
|
|
ABC-BPNN
|
0.8998 (+0.0564)
|
|
SSA-BPNN
|
0.9233 (+0.0799)
|
|
SOA-BPNN
|
0.9545 (+0.1111)
|
|
ISOA-BPNN
|
0.9762 (+0.1328)
|
The accuracy obtained was 0.8434 when the original BPNN algorithm was used to predict
the mental health status of college students. After optimizing the BPNN parameters
with the optimization algorithm, several algorithms showed varying degrees of improvement
in their predictive performance compared to the BPNN algorithm. The PSO-BPNN algorithm
achieved an accuracy of 0.8872, indicating 0.0438, 0.0564, and 0.0799 increases compared
to the BPNN algorithm, ABC-BPNN algorithm, and SSA-BPNN algorithm, respectively. The
accuracy of the SOA-BPNN algorithm was 0.9545, showing a 0.1111 increase compared
to the BPNN algorithm. This result suggests the advantage of SOA in optimizing the
parameters of the BPNN algorithm. The ISOA-BPNN algorithm designed in this paper demonstrated
an accuracy of 0.9762, which was 0.1328 and 0.0217 higher than the BPNN and SOA-BPNN
algorithms, respectively. This result verified the reliability of the improvement
of SOA. Table 10 compares the F1 values and AUCs.
Table 10. Improvement of BPNN in terms of the F1 Value and AUC by Different Optimization Algorithms.
|
|
F1 value
|
AUC
|
|
BPNN
|
0.8991
|
0.9328
|
|
PSO-BPNN
|
0.9322 (+0.0331)
|
0.9633 (+0.0305)
|
|
ABC-BPNN
|
0.9457 (+0.0466)
|
0.9726 (+0.0398)
|
|
SSA-BPNN
|
0.9528 (+0.0537)
|
0.9876 (+0.0548)
|
|
SOA-BPNN
|
0.9721 (+0.0730)
|
0.9901 (+0.0573)
|
|
ISOA-BPNN
|
0.9834 (+0.0843)
|
0.9956 (+0.0628)
|
The original BPNN algorithm had an F1 value of 0.8991 and an AUC of 0.9328. After
algorithm optimization, the F1 value and the AUC exhibited substantial improvements.
In particular, the F1 value of the ISOA-BPNN algorithm was 0.9834, showing a 0.0843
increase compared to the BPNN algorithm, while the AUC was 0.9956, signifying a 0.0628
increase. Hence, the ISOA algorithm is effective in optimizing the BPNN parameters,
and the optimized algorithm has superior capability to differentiate the mental health
status of college students.
The ISOA-BPNN model was then compared with other predictive models, and the statistical
difference was analyzed. The comparison models included the following:
(1) logistics regression [20],
(2) support vector machine (SVM) [21],
(3) XGBoost [22].
The results are presented in Table 11.
The logistic regression model performed significantly worse than the other models
(Table 11). When predicting the mental health status of college students, the F1 and AUC values
were only 0.9126 and 0.9532, respectively, the lowest among all the models compared.
The difference was statistically significant compared to the ISOA-BPNN model (p <
0.01). The F1 and AUC values of the SVM and XGBoost models were above 0.95, but they
were still significantly lower than the ISOA-BPNN model (p < 0.05), demonstrating
the superiority of the proposed method for predicting the mental health status of
college students. The proposed model could accurately identify college students who
may experience anxiety, depression, and other negative emotions.
Table 11. Comparison of F1 and AUC Values between Different Models.
|
|
F1 value
|
AUC
|
|
Logistics regression
|
0.9126±0.0125**
|
0.9532±0.0126**
|
|
SVM
|
0.9354±0.0133*
|
0.9627±0.0108*
|
|
XGBoost
|
0.9577±0.0121*
|
0.9889±0.0132*
|
|
ISOA-BPNN
|
0.9834±0.0145
|
0.9956±0.0137
|
Note: **: compared to the ISOA-BPNN model, p < 0.01; *: compared to the ISOA-BPNN
model, p < 0.05.
Finally, ten samples were analyzed to predict the mental health status of college
students, as listed in Table 12.
Across the five samples under study, the BPNN outputs aligned perfectly with the actual
conditions, affirming the reliability of the ISOA-BPNN algorithm in predicting the
mental health status of college students, as shown in Table 12. Of the five samples, three were categorized as normal, while two were considered
abnormal. The two abnormal samples were analyzed. Sample 2 showed heavy stress on
items X5, X6, X7, and X8. Further investigation suggested that the student had poor
exam performance, experienced financial difficulties, and faced criticism from a teacher
due to classroom performance. These factors contributed to a high SCL-90 score. Such
a student requires psychological counseling for relief. Sample 5 displayed elevated
stress on items X1, X2, and X3. Through a practical inquiry, it was discovered that
this senior student faced increased pressure related to employment, leading to heightened
emotional distress, anxiety, and depression. Hence, the school should focus on providing
the student with meaningful employment guidance to help alleviate these negative emotions.
Table 12. Predictive Analysis of the Mental Health Status of College Students.
|
Sample number
|
ISOA-BPNN input
|
Actual result
|
ISOA-BPNN output
|
|
X1
|
X2
|
X3
|
X4
|
X5
|
X6
|
X7
|
X8
|
X9
|
X10
|
|
1
|
1
|
1
|
1
|
2
|
1
|
1
|
1
|
1
|
1
|
1
|
Normal
|
Normal
|
|
2
|
3
|
3
|
3
|
3
|
4
|
4
|
4
|
4
|
2
|
2
|
Abnormal
|
Abnormal
|
|
3
|
1
|
1
|
1
|
1
|
1
|
1
|
1
|
1
|
1
|
1
|
Normal
|
Normal
|
|
4
|
2
|
2
|
1
|
1
|
1
|
1
|
2
|
1
|
1
|
1
|
Normal
|
Normal
|
|
5
|
4
|
4
|
4
|
3
|
3
|
3
|
3
|
3
|
2
|
2
|
Abnormal
|
Abnormal
|
5. Discussion
The psychological well-being of college students has always been a focal point in
psychology and education. With the continuous development of big data technology,
an increasing number of big data methods are being applied in the mental health field,
revealing psychological characteristics more comprehensively and objectively. The
article primarily used big data technology to comprehend the stress and mental health
status of college students through questionnaires. It integrated information from
various aspects, such as employment and academic performance, to comprehensively predict
college students‘ mental health status.
The ISOA algorithm outperformed the PSO, ABC, and the other optimization algorithms.
The findings from the evaluation of five benchmark test functions demonstrated that
the ISOA exhibited superior optimization ability and achieved better parameter optimization
performance for BPNN. Tables 9 and 10 also confirmed this point, as the ISOA-optimized BPNN model outperformed in
terms of accuracy, F1 value, and AUC value for predicting the mental health status
of college students. Hence, the enhancement effect of the ISOA on the predictive performance
of BPNN was demonstrated, further improving its effectiveness in predicting the mental
health status of college students. Furthermore, The logistic regression and other
methods yielded significantly lower F1 and AUC values than the ISOA-BPNN model (p
< 0.05), highlighting the advantages of the proposed ISOA-BPNN approach.
The ISOA-BPNN model accurately determined the psychological health status of college
students, providing reliable support for guidance and intervention in schools. This
method can be implemented in real universities to improve the mental health of college
students and promote social stability. Nevertheless, this study also has some limitations.
Data privacy protection and potentially biased data were not considered. Moreover,
the research was conducted solely on universities located in Hebei province. Future
research will explore how to achieve a better balance between data privacy and data
mining while examining a broader range of data to determine the applicability of the
proposed methods.
6. Conclusion
Research was conducted on predicting the psychological well-being of college students.
A questionnaire survey was used to understand the stress levels and mental health
conditions of college students. Subsequently, an ISOA-BPNN predictive model was developed
to forecast the psychological well-being of college students. Compared to other optimization
algorithms, the ISOA had better optimization ability and was more effective in enhancing
the performance of the BPNN algorithm for predicting mental health. The ISOA-BPNN
model achieved an F1 value and an AUC of 0.9834 and 0.9956, respectively, in predicting
the mental health of college students. The proposed model had a clear advantage over
other models, such as the SVM. These results affirm the reliability of the optimized
method and its potential for practical application and broader adoption. This article
demonstrates the superior optimization effect of the ISOA and provides a new and reliable
method for parameter optimization in a BPNN. It also offers theoretical references
for further research on optimization problems.
REFERENCES
X. Wang, ``Research on Mental Health Education for College Students,'' International
Journal of Social Science and Education Research, Vol. 3, No. 3, pp. 153-157, 2020.
E. E. Fitzsimmons-Craft, C. B. Taylor, M. G. Newman, N. H. Zainal, E. E. Rojas-Ashe,
S. K. Lipson, M. L. Firebaugh, P. Ceglarek, К. Койнова, N. Topooco, N. C. Jacobson,
A. K. Graham, H. M. Kim, D. Eisenberg, and D. E. Wilfley, ``Harnessing mobile technology
to reduce mental health disorders in college populations: A randomized controlled
trial study protocol,'' Contemporary Clinical Trials, Vol. 103, pp. 1-10, 2021.
M. Gibson, N. Anderson, A. Horton, K. Mullions, and A. Reynolds, ``0255 Sleep Hygiene
Index: Associations with Sleep and Mental Health in College Students,'' SLEEP, Vol.
45, No. Supplement_1, pp. A114-A115, 2022.
S. P. Becker, A. S. Holdaway, and A. M. Luebbe, ``Suicidal Behaviors in College Students:
Frequency, Sex Differences, and Mental Health Correlates Including Sluggish Cognitive
Tempo,'' Journal of Adolescent Health, Vol. 63, pp. 181-188, 2018.
W. J. B. W. Hamat, H. M. B. Ishak, K. Hashikura, T. Suzuki, and K. Yamda, ``Detection
of Anxiety Expression From EEG Analysis Using Support Vector Machine,'' ECTI Transactions
on Electrical Engineering, Electronics, and Communications, Vol. 17, No. 1, pp. 87-94,
2019.
Y. Tan, Q. Zeng, and H. Zhang, ``Research on anxiety detection based on personalized
data markers,'' Journal of Physics: Conference Series, Vol. 1948, No. 1, pp. 1-9,
2021.
H. Akbari, M. T. Sadiq, M. Payan, S. S. Esmaili, H. Baghri, and B. Hamed, ``Depression
Detection Based on Geometrical Features Extracted from SODP Shape of EEG Signals and
Binary PSO,'' Traitement du Signal, Vol. 38, No. 1, pp. 13-26, 2021.
S. Sardari, B. Nakisa, M. N. Rastgoo, and P. Eklund, ``Audio based depression detection
using Convolutional Autoencoder,'' Expert Systems with Applications, Vol. 189, pp.
1-13, 2022.
Y. Fauziah, S. Saifullah, and A. S. Aribowo, ``Design Text Mining for Anxiety Detection
using Machine Learning based-on Social Media Data during COVID-19 pandemic,'' Proceeding
of LPPM UPN ``Veteran'' Yogyakarta Conference Series 2020 - Engineering and Science
Series, Vol. 1, No. 1, pp. 253-261, 2020.
M. Papadakaki, S. Lioliou, V. Kounali, A. Liangou, and J. Chliaoutakis, ``Mental health
burden and self-efficacy among undergraduate students in the health sciences,'' European
Journal of Public Health, Vol. 31, No. Supplement_3, 2021.
L. R. Derogatis, R. S. Lipman, K. Rickels, E. H. Uhlenhuth, and L. Covi, ``The Hopkins
Symptom Checklist (HSCL): A self-report symptom inventory,'' Behavioral Science, Vol.
19, No. 1, pp. 1-15, 1974.
Y. Wang, and G. Sun, ``Oversampling method for intrusion detection based on clustering
and instance hardness,'' Journal of Computer Applications, Vol. 41, No. 6, pp. 1709-1714,
2021.
X. Liu, X. Zhao, and W. B. Shangguan, ``Fatigue life prediction of natural rubber
components using an artificial neural network,'' Fatigue & Fracture of Engineering
Materials and Structures, Vol. 45, No. 6, pp. 1678-1689, 2022.
M. Huang, C. Hong, J. Chen, C. Ma, C. Li, and Y. Huang, ``Prediction of Peak Shear
Strength of Rock Joints Based on Back-Propagation Neural Network,'' International
Journal of Geomechanics, Vol. 21, No. 6, pp. 1-11, 2021.
N. Panagant, N. Pholdee, S. Bureerat, K. Kaen, A. R. Yıldız, and S. M. Sait, “Seagull
optimization algorithm for solving real-world design optimization problems,” Materials
Testing, Vol. 62, No. 6, pp. 640-644, 2020.
F. Liu, Y. Sun, G. Wang, and T. Wu, “An Artificial Bee Colony Algorithm Based on Dynamic
Penalty and Lévy Flight for Constrained Optimization Problems,” Arabian Journal for
Science & Engineering, Vol. 43, No. 12, pp. 1-20, 2018.
H. Yu, M. Zheng, W. Zhang, W. Nie, and T. Bian, ``Optimal design of helical flute
of irregular tooth end milling cutter based on particle swarm optimization algorithm,''
Proceedings of the Institution of Mechanical Engineers, Part C. Journal of Mechanical
Engineering Science, Vol. 236, No. 7, pp. 3323-3339, 2022.
S. Singh, N. J. Singh, and A. Gupta, ``System sizing of hybrid solar‐fuel cell battery
energy system using artificial bee colony algorithm with predator effect,'' International
Journal of Energy Research, Vol. 46, No. 5, pp. 5847-5863, 2022.
G. F. Fan, Y. Li, X. Y. Zhang, Y. H. Yeh, and W. C. Hong, ``Short‐term load forecasting
based on a generalized regression neural network optimized by an improved sparrow
search algorithm using the empirical wavelet decomposition method,'' Energy Science
and Engineering, Vol. 11, No. 7, pp. 2444-2468, 2023.
R. Xie, R. Liu, X. B. Liu, J. M. Zhu, and G. Mustafa, ``Evaluation of SMEs’ Credit
Decision Based on Support Vector Machine-Logistics Regression,'' Journal of Mathematics,
Vol. 2021, No. 8, pp. 1-10, 2021.
F. Camastra, V. Capone, A. Ciaramella, A. Riccio, and A. Staiano, ``Prediction of
environmental missing data time series by Support Vector Machine Regression and Correlation
Dimension estimation,'' Environmental Modelling & Software, Vol. 150, pp. 1-7, 2022.
Z. Zhang, Y. Huang, R. Qin, W. Ren, and G. Wen, ``XGBoost-based online prediction
of seam tensile strength for Al-Li alloy in laser welding: Experiment study and modelling,''
Journal of Manufacturing Processes, Vol. 64, No. May, pp. 30-44, 2021.
Author
Peng Zhang was born in Cangzhou, Hebei, China, in 1988. From 2011 to 2014, he studied
at Tianjin Normal University and received his Master’s degree in 2014. Currently,
he works at Cangzhou Normal University. He has published eight papers and has presided
over six projects at the provincial and municipal levels. His research interests are
the ideological and political education of college students.
Wenjing Han was born in Hengshui, Hebei Province, in 1973. She studied at the Department
of Education at Hebei University from 1993 to 1997 and gained a Bachelor’s degree.
She began work in 1997. From 2007 to 2009, she studied in the postgraduate program
for college and university teachers at Hebei University and gained a Master’s degree.
At present, she is a lecturer at Handan University. In recent years, she has mainly
been researching positive psychology. She has presided over eight provincial research
projects and published several articles.
Quanzhi Liu was born in Cangzhou, Hebei, China, in 1987. From 2012 to 2015, he studied
in Hebei University and received his Master’s degree in 2015. Currently, he works
in Cangzhou Normal University. He has published several papers and has presided over
five projects at the municipal level. His research interests are statistical analysis
and financial management.