Statistical Insights into the Influence of Music on Mental Health

Xinyi Zhang

School of Management, Fudan University, Shanghai, 200433, China

Keywords: Logistic Regression, Mental Health, Music, Machine Learning.

Abstract: Mental Health has recently become a heated topic among the public, ranging from biologists seeking for

pathological cause to psychologists who attempts to improve the dilemma. In this paper, a comprehensive set

of methods such as descriptive analysis, logistic regression, random forest and XGBoost will be adopted to

analyse the dataset from Kaggle and establish the relationship between music and mental health as well, with

an aim to provide some evidence on the effect of music therapy, an emerging curing method. By comparing

the figure of the Root Mean Squared Error (RMSE), it can be concluded that logistic regression performs the

best among these models in predicting mental health states. Nevertheless, the actual figures of RMSE still

suggest room for improvement. For future related experiments, significant improvements can be done by

increasing the volume of samples, perfecting data filtration, adding more variables and further refining

parameters of the model. After all, this research provides empirical evidence for the application of music

therapy by using statistical models while at the same time offering keen insights for further improvements.

1 INTRODUCTION

In recent years, the public’s awareness of personal

well-being is gradually increasing, frequently

accepting the transformation to healthy lifestyles.

While physical health has remained necessary, as

evidenced by growing attention paid on exercise and

nutrition, it is generally acknowledged that

psychological health issues need to be highly valued

as well. Among all the people agonizing over mental

problems, it is the adolescents that suffer the most.

The World Health Organization (WHO) reported in

2021 that one in seven individuals aged between 10

to 19 suffered from a mental disorder. In addition,

statistics indicate that there is a rising prevalence of

mental health issues in the current generation of

adolescents (often referred to as “Generation Z”)

falling victim to a “mental health crisis” (Borg et al.,

2024). Contributing factors proposed range from

social media misuse to enhanced tension arisen from

adapting to novel academic situations (Chen, 2024).

There are varied types of psychological disorders

such as depression, anxiety, somatization and

interpersonal sensitivity (Wang, 2023). These

symptoms partly account for poor academic

performance, severe insomnia and even disability in

https://orcid.org/0009-0002-7040-5647

daily activities (Yang et al., 2025). Given the severe

consequences, addressing adolescents’ mental health

has become researchers’ critical focus.

Currently, researchers have explored various

effective strategies to alleviate psychological issue

suffering, with workshops and psychological

counselling been more frequently implemented from

primary schools to universities. Scientists also

proposed the possibility to use artificial intelligence

models to predict psychiatric risk with an aim of

better precaution (Hill et al., 2025). Besides this,

based on previously published samples, psychologists

have established Positive Mental Health (PMH) Scale

for middle school students, enabling teenagers to

evaluate their mental states by themselves (Han et al.,

2025).

Among all the emerging therapeutic approaches,

music therapy, which differs from traditional curing

methods, has gained attention among both musicians

and psychologists. To be more specific, music

composition and group music-making, music

medication and music-based self-care are all

considered effective intervention strategies (He et al.,

2024). The preliminary practice of music therapy is

initiated in several colleges. Students who participate

in the practice are given tailored immersive

Zhang, X.

Statistical Insights into the Inﬂuence of Music on Mental Health.

DOI: 10.5220/0013826900004708

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 2nd International Conference on Innovations in Applied Mathematics, Physics, and Astronomy (IAMPA 2025), pages 413-418

ISBN: 978-989-758-774-0

413

experience according to their psychological state

assessment in advance (Hu, 2024). Biology

researchers also give insights of the mechanism of

music therapy. Furthermore, studies have identified

that pop fans exhibit fewer depressive symptoms,

while all-out rock fans behaved the worst in personal

well-being (Bogt et al., 2021). Based on the vocal

communication theory (VST), humans perceive tones

in different ways depending on the behavioral

significance in vocal communication (Bowling,

2023). This theory provided biological basis for

applying specific music to patients with different

symptoms.

While a majority of existing studies merely focus

on case studies and additional biological theories, this

research will provide empirical evidence in the

correlation between exposure to music and

adolescents’ mental health states through big data and

statistical methods. This research aims to provide

empirical evidence on the influence of different types

of music on adolescents’ mental health through the

use of statistics. Methods of the study include

descriptive analysis, correlation analysis, decision

tree and multiple linear regression. In addition, the

study will further compare the performance by

indexes such as confusing matrix and mean squared

error (MSE). Data used in the study is found on

Kaggle. By quantifying the association between

music exposure and psychology states, the study

seeks to establish some scientific evidence for the

application of music therapy in adolescents’ mental

disorder interventions.

2 METHODOLOGY

2.1 Data Source

The data utilized in the research is obtained from the

Kaggle website, specifically the “Music & Mental

Health Survey Results”. The data collection process

was carried out via a Google Form in 2022, with no

restrictions of respondents’ age or location.

Furthermore, it has been downloaded around 31.7

thousand times and has won the usability score of

10.00, indicating high data quality. The dataset is in

.csv format.

From the overall dataset, a variety of factors

relating to music and mental health can be observed.

Such diversity in variable selection facilitates

comprehensive analysis in the research.

2.2 Variable Selection and Description

This dataset as a whole contains observations and 33

variables. In order to clarify the influence of music on

individuals’ mental health, irrelevant columns such as

“timestamp” and “Permissions” are filtered out. For

data analysis, there will eventually be 31 columns.

After further filtering steps, 12 out of all 31 variables

are the most relevant, clustered to two types,

“Numeric” and “Categorial”, and the terms and

explanations of each variable are shown in Table 1.

Table 1: Variable Description

Ter

e Ex

lanation

e Numeric The a

e of res

ondents, ran

from10 to 25

Hours per da

Numeric Hours per day spent in listening to music

While workin

Cate

orial Whether res

ondents listen to music while workin

, 1 for

es, 0 for no

Instrumentalist Categorial Whether respondents are able to play an instrument, 1 for yes, 0 for no

Com

ose

Cate

orial Whether res

ondents can com

ose music themselves, 1 for

es, 0 for no

Fav genre

BPM

Anxiety

Depression

Insomnia

OCD

Music effects

Categorial

Numeric

Categorial

Respondents’ favorite or top genre

Beats per minute of respondents’ favorite genres

Self-reported anxiety, on a scale of 0-10

Self-reported depression, on a scale of 0-10

Self-reported insomnia, on a scale of 0-10

Self-reported OCS, on a scale of 0-10

Whether music improve or worsen respondents’ mental health

conditions, 1 for im

rove, 0 for no effect, -1 for worsen

IAMPA 2025 - The International Conference on Innovations in Applied Mathematics, Physics, and Astronomy

414

Graphs indicating the features of the dataset are

shown below.

Figure 1: The Histogram of Age Distribution (Picture

credit: original)

From figure 1, it can be concluded that the average

age of the respondents is 25.21, the maximum and

minimum figure of which is 89 and 10 respectively.

The graph depicts a right skewed distribution and

respondents with an age of 18 have the most counts.

Figure 2: Pie Chart of Whether Respondents Play an

Instrument (Picture credit: original)

Figure 3: Pie Chart of Whether Respondents Compose

Music (Picture credit: original)

Figure 4: Pie Chart of Whether Respondents Listen to

Music While Working (Picture credit: original)

Figure 2 indicates that 32.6% of the interviewees play

an instrument besides listening to music. Only a slight

17.5% of them can compose music themselves,

according to figure 3. However, as indicated by figure

4, a high percentage of 79.7% claim that they listen to

music while working.

Statistical Insights into the Inﬂuence of Music on Mental Health

415

Figure 5: Heatmaps of Music Genres and Mental Health States of depression (Picture credit: original)

Additionally, the color gradient of the heatmap in

figure 5 indicates different figures regarding the

match between favorite and level of mental health.

Nevertheless, there are no clear and strong patterns

uniformly emerging in four types of mental health

issues. A more complicated relationship needs to be

determined.

2.3 Methods Introduction

The paper combines statistical methods and machine

learning methods, including descriptive analysis,

logistic regression, random forests and decision tree.

Firstly, descriptive analysis provides an elementary

summary of the whole dataset and visualizes variable

features. For numerical variables such as “age”,

histograms are adopted to provide data distribution.

For binary and multiple categorial variables such as

“While working”, “Primary streaming service”, etc.,

pie charts are adopted. Additionally, heatmaps

explore the relationship between two variables.

Secondly, logistic regression model is utilized to

predict the influence of music-related indexes on

different mental health conditions, given a certain

trained dataset and a test dataset with a proportion of

8:2. To facilitate logistic regression, the data related

to the level of different mental health states are

divided into two groups, 1 for the figure over 4, and 0

for the else. The article used confusion matrix to

present figures of false positive (type I error) and false

negative (type II errors) and the accuracy of the

logistic regression model. Thirdly, the random forest

IAMPA 2025 - The International Conference on Innovations in Applied Mathematics, Physics, and Astronomy

416

classifier builds multiple decision trees to determine

feature importance. XGBoost, a gradient-boosting

library, iteratively improve the model by fitting the

residuals of previous ones. In every iteration, the

model adopts decision tree as its basic learners.

Generally, the three methods’ performance is

assessed using RMSE to propose a better strategy for

a specific dataset. R and Python are employed for

programming.

3 RESULTS AND DISCUSSION

3.1 Model Results

The visualized results of confusion matrix using

logistic regression and XGBoost and the graph

indicating feature importance are depicted below.

Figure 6: Confusion Matrix of Logistic Regression

Regarding Anxiety (Picture credit: original)

Figure 7: Confusion Matrix of Logistic Regression and

XGBoost Regarding Anxiety (Picture credit: original)

From figure 6 and 7, for anxiety prediction, the

number of samples with a true label of 0 but predicted

label of 1 is 39, and the number of samples with a true

label of 1 and a predicted label of 0 is 0. The figure of

correctly predicted cases is 85. On the other hand, the

XGBoost model possesses 63 cases of true positives,

8 false negatives, 22 true negatives and 31 false

positives. It can be concluded that XGBoost performs

overall better than logistic regression, but stull with

an imperfect performance.

Cases of true positives, false negatives, true

negatives and false positives regarding logistic

regression model for the other three type of mental

health statuses are (59,9,10,46), (8,63,42,11),

(1,91,31,1) respectively, as shown in figure 6, figure

7 and figure 8. Those for XGBoost are (36,20,33,35),

(14,57,36,17), (1,75,31,17) respectively.

Figure 8: Feature Importance of Random Forest (Picture credit: original)

Statistical Insights into the Inﬂuence of Music on Mental Health

417

Figure 8 is the feature importance graph of random

forest depicts that “BPM” (beats per minute) occupies

the highest importance in affecting music listeners’

mental health states. “Fav genre” (favorite genre)

ranks the second and “Hours per day” has a moderate

effect. “Music effects” (self-analyzed music effects

on individual mental health states) contributes the

least to predicting mental health conditions using

random forest.

3.2 Compare Results

Based on table 2, which shows the RMSE value for

each model and for each category of mental health

issues, logistic regression generally outperforms

random forest and XGBoost since it has relatively

low RMSEs among all four mental health states,

indicating a more accurate prediction ability.

Nevertheless, the logistic regression model is never

without room for improvement.

Table 2: RMSE values for all three models regarding four mental health states

RMSE Anxiet

ression Insomnia OCD

Logistic Regression 0.5608 0.6720 0.6538 0.5080

Random Forest 0.6720 0.7184 0.6660 0.5820

XGBoost 0.6538 0.7405 0.6538 0.6222

4 CONCLUSION

In conclusion, although the percentages of false

predictions are not so significant in terms of both

logistic regression and XGBoost, the RMSE figure is

not good enough. Overall, the paper partly

demonstrates the influence of music on mental health,

with beats per minute (BPM) affecting the most.

Nevertheless, the research might be more convincing

if the RMSE values are lower than 0.1. For future

improvement, the reason of the relatively bad

performance must be determined. Poor data quality is

the culprit of underperformance since the self-

assessment of mental health indicators might deviate

due to subjective factors and inconsistent criteria,

which will affect the effectiveness of the dataset. Also

the lack of restrictions in data collection may also lead

to some noise. In addition, the model itself should be

modified by adjusting the parameters and considering

probable other complicated relationship between

music and mental health states rather than splitting

the level into two groups. Last but not least, other

variables such as the time period of listening to music

in a day are not taken into consideration. These

factors may also affect people’s mental health status.

In conclusion, while the performance of the proposed

model does not meet researcher’s expectation, it does

provide some insights into the relationship between

music and mental health, and at the same time paves

the way for future optimization.

REFERENCES

Bogt, T. T., Hale, W. W. Becht, A. 2021. Wild Years: Rock

Music, Problem Behaviors and Mental Well-being in

Adolescence and Young Adulthood. J Youth, 50, 2487-

2500.

Borg, M. E., Heffer, T. Willoughby, T. 2024. Generational

Shifts in Adolescent Mental Health: A Longitudinal

Time-Lag Study. J. Youth Adolescence, 1-12.

Bowling, D. L. 2023. Biological Principles for Music and

Mental Health. Transl Psychiatry, 13, 374.

Chen, L. F. 2024. Analysis on the Construction Path of

College Students' Mental Health Education and

Guidance Mechanism in the New Media Era. News

Research Guide, 15 (01), 148-150.

Han, Y., Yang, Y., Shentu, M. L., et al. 2025. Positive

Mental Health (PMH) Scale for Middle School

Students: Structure and Measure. Journal of Peking

University (Natural Science), 61(01), 166-172.

He, Q., Attan, S. A., Zhang, J. et al. 2024. Evaluating Music

Education Interventions for Mental Health in Chinese

University Student: a Dual Fuzzy Analytic Method. Sci

Rep, 14, 19727.

Hill, E. D., Kashyap, P., Raffanello, E. et al. 2025.

Prediction of Mental Health Risk in Adolescents. Nat

Med, 1-7.

Hu, P. Y. 2024. Application of Music Therapy in College

Students' Mental Health Education. Education Science

& Culture Magazine, 18, 184-188.

Wang, L. 2023. Exploring the Influence of Music

Education on the Development of College Mental

Health based on Big Data. Soft Comput. 27, 17213-

17229.

Yang, Y. Q., Huang, J., Lin, Z. X., et al. 2025. Mental

Health Literary and its Relationship with Symptoms of

Depression, Anxiety, and Insomnia in Young Adults.

Chinese Mental Health Journal, 4, 344-349.

IAMPA 2025 - The International Conference on Innovations in Applied Mathematics, Physics, and Astronomy

418