Determinants of University Students’ Sleep Quality

Xixiang Gaoqiao

Department of Mathematics, Durham University, Durham, DH1 3LE, U.K.

Keywords: Sleep Quality, Behavioural Factors, Predictive Modelling, Linear Regression, Random Forest.

Abstract: University students often experience poor sleep quality due to various academic and lifestyle pressures. This

study analyzes a dataset of 500 university students to identify key determinants of sleep quality. This paper

examines demographic factors (age, gender, year of study), daily habits (study hours, screen time, caffeine

intake, physical activity), and sleep characteristics (sleep duration, bed/wake times). Multiple linear regression

and random forest models were used to explore linear and non-linear relationships. The linear regression

models showed very low explanatory power in both the full dataset (n = 500) and a filtered dataset (n = 52),

suggesting that none of the measured factors had a significant linear impact. The random forest models

performed well on the training set but had poor performance on the test set, indicating possible overfitting.

However, both models consistently highlighted sleep duration and study hours as relatively important features.

These findings imply that unmeasured factors, such as psychological stress, may play a larger role in students’

sleep quality. In conclusion, while lifestyle and demographic variables alone did not strongly predict sleep

quality in this sample, the results highlight the need to consider psychological and environmental factors. This

paper discusses the implications, limitations (including the lack of psychological/cultural variables and

seasonal effects), and provide suggestions for future research and interventions to improve student sleep

health.

1 INTRODUCTION

Sleep is essential for students’ health and academic

success. However, many university students do not

sleep well due to irregular schedules, heavy study

loads and social activities. Surveys have reported that

70.6% of university students get less sleep than the

recommended 8 hours per night (Hershner and

Chervin, 2014). Some studies even found that up to

31–65% of students suffer from poor sleep quality

(Zhang et al., 2022). This lack of sleep can harm their

learning and health. For example, insufficient sleep

leads to worse academic performance and serious

physical health outcomes. It is also associated with

mental health conditions like depression and anxiety

(Zhang et al., 2022). In short, poor sleep among

students is common and it has negative consequences

for multiple aspects of their daily lives.

Understanding which factors have the greatest

influence on sleep quality is important for designing

effective interventions and health policies.

https://orcid.org/0009-0001-4307-7716

The aim of this paper is to investigate the

determinants of university students’ sleep quality

using a comprehensive dataset. The dataset is fetched

from the Kaggle website (Student Sleep patterns)

which contains 500 groups of data without any

missing data. Each student record contained

demographic information (age, gender, and year of

study), behavioural factors (average study hours per

day, daily screen time, caffeine intake, and weekly

physical activity), and sleep-related variables

(average sleep duration in hours, a self-reported sleep

quality score from 1 to 10, and typical sleep/wake

times on weekdays and weekends). In addition, two

new variables-Weekday Sleep Duration and

Weekend Sleep Duration-were created to better

reflect sleep patterns. A filtered dataset (n = 52) was

also prepared by removing records with unrealistic or

inconsistent sleep schedule data. This paper examines

how demographic characteristics and daily habits

relate to students’ self-rated sleep quality. Multiple

linear regression and random forest models were

applied to identify key predictors. The results aim to

clarify which factors are most influential for student

Gaoqiao, X.

Determinants of University Students’ Sleep Quality.

DOI: 10.5220/0013848000004708

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 631-637

ISBN: 978-989-758-774-0

631

sleep and to provide insights that may guide future

interventions. The following sections present a

review of relevant literature, the methodology of the

analysis, the results obtained, and a discussion of

conclusions, limitations, and future outlook.

2 RELATED WORKS

The topic of students’ sleep quality has received a lot

of interest in recent years. For instance, Lund et al.

conducted a large survey of college students in the

United States and found that over 60% of students

were categorized as poor-quality sleepers according

to the Pittsburgh Sleep Quality Index (PSQI). In that

study, students commonly had delayed bedtimes on

weekends and reported that stress from academics

and emotions negatively affected their sleep (Lund et

al., 2010; Yang et al., 2012).

Moreover, Becker et al. surveyed over 7,600

students across multiple universities and found that

62% met the criteria for poor sleep quality.

Interestingly, they observed some gender differences

(with female students reporting slightly higher rates

of sleep problems than males) and found that mental

health symptoms were strongly associated with sleep

issues (Becker et al., 2018). Similarly, a study in

Ethiopia by Lemma et al. reported 55.8% of

university students as having poor sleep quality

(PSQI > 5). That study found that female students and

those in higher years of study had higher odds of poor

sleep, and importantly, higher perceived stress and

symptoms of depression/anxiety were strongly

correlated with worse sleep quality (Lemma et al.,

2014).

Finally, recent research has started to examine

other potential determinants of student sleep quality

beyond psychological stress. Schmickler et al.

conducted a cross-sectional study with university

students in Germany to identify what influences sleep

quality. They found nearly half of the students

(48.7%) had poor sleep quality (PSQI > 5). Their

regression analysis indicated that several factors

significantly predicted poor sleep: older students had

worse sleep, and students who reported higher stress

and exhaustion levels also had lower sleep quality

(Schmickler et al., 2023).

Most of these studies used regression analysis or

basic statistical tests to find significant factors.

However, fewer studies have compared traditional

models with machine learning methods to see if

prediction can be improved. Overall, previous

literature suggests that poor sleep is common among

students especially age and gender, mental health and

stress are often the main predictors. However, fewer

studies have examined whether common daily habits

(such as screen time, caffeine use, or exercise) can

predict sleep quality in a more detailed model.

3 METHODOLOGIES

3.1 Data Source

A dataset of 500 university students is used. The

original dataset remained in the .CVS format. The key

variables are collected in Table 1:

Table 1: Descriptions of variables used in the dataset.

Variables Symbol Description

Sleep Quality X1 The outcome variable. This was

a self-reported rating of overall

sleep quality on a scale from 1

(very poor) to 10 (excellent).

Age X2 The student’s age in years.

Gender X3 Categorical variable with three

groups: Female, Male, an

Other.

University

Year

X4 The student's current year o

university (categorical: '1st

Year', '2nd Year', '3rd Year', '4th

Year').

Sleep

Duration

X5 The average number of hours the

student sleeps per night.

Study Hours X6 The average number of hours per

day the student spends on

academic work.

Screen Time X7 The average hours per day the

student spends on screens.

Caffeine

Intake

X8 The typical number o

caffeinated beverages the studen

consumes in a day.

Physical

Activity

X9 The student’s level of physical

activity, measured by minutes o

exercise per week.

Weekday

Bedtime

X10 The typical time the student goes

to bed on weekdays reported i

hour of day.

Weekday

Wake-up

Time

X11 The typical time the student

wakes up on weekdays.

Weekend

Bedtime

X12 The typical bedtime on

weekends.

Weekend

Wake-up

Time

X13 The typical wake-up time o

weekends.

All time-of-day variables (bedtimes and wake times)

were recorded in hours using a 24-hour format with

decimals. These times allow people to capture

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

632

differences in sleep schedule between weekdays and

weekends. Before analysis, this paper checked and

cleaned the data for any inconsistencies or data loss.

In addition to the original variables, two new

variables were created to better capture students’

sleep patterns: Weekday Sleep Duration and

Weekend Sleep Duration. These were calculated by

subtracting bedtime from wake-up time separately for

weekdays and weekends. Durations less than zero

were adjusted by adding 24 to reflect overnight sleep

accurately. To ensure that the calculated sleep

durations were meaningful and reflected typical sleep

behaviour, specific filtering criteria were applied.

Bedtimes were required to fall between 19:00 (7 PM)

and 5:00 (5 AM), and wake-up times were required to

be between 4:00 (4 AM) and 11:00 (11 AM). If a

student's bedtime or wake-up time fell outside these

ranges, the corresponding sleep duration was

considered invalid and marked as missing (Table 2).

Table 2: Descriptions of new variables.

Variables Symbol Description

Weekday

Sleep

Duration

X14 Hours of sleep on a typical

weekday night (calculated as

Wake - Bedtime)

Weekend

Sleep

Duration

X15 Hours of sleep on a typical

weekend night (calculated as

Wake - Bedtime

)

After applying this filtering process, only the records

with valid Weekday Sleep Duration and Weekend

Sleep Duration remained in the dataset. This resulted

in a filtered dataset of 52 students, which was used in

a separate analysis to compare with the full dataset of

500 students. The full dataset allowed a broader

analysis using only the original Sleep Duration

variable, while the filtered dataset provided more

precise measurements of sleep patterns, enabling

deeper insights into weekday and weekend sleep

behaviour (Sohn et al., 2012).

For clarity, all independent variables are labeled

as X1 to X15, as shown in Table 1 and in Table 2,

which lists their corresponding names and

descriptions.

Categorical variables, including Gender (X3) and

University Year (X4), were transformed into dummy

variables for use in the models. For Gender, three

variables were labeled: Male (X3_1), Other (X3_2),

and Female (X3_3). Although Female is the reference

category in the regression models and does not have

a coefficient, it is still assigned a code (X3_3) for

clarity and consistency in tables. Similarly, for

University Year, four variables were labeled: 1st year

(X4_1), 2nd year (X4_2), 3rd year (X4_3), and 4th

year (X4_4), where 1st year serves as the reference

group but is also coded for consistency.

Two main analytical approaches are employed to

investigate the determinants of sleep quality. Data

analysis was conducted with the help of SPSSAU, an

online application for statistical modeling.

3.2 Linear Regression

First, a multiple linear regression model is built with

Sleep Quality as the dependent variable. All the other

measured factors (age, gender, university year, sleep

duration, study hours, screen time, caffeine intake,

physical activity, and the sleep schedule variables)

were used as predictors. Before fitting the model,

categorical variables (Gender and Year of Study)

were encoded into dummy variables. For example,

Gender was represented with two dummy indicators

(Male and Other, with Female as the reference

category), and Year of Study was represented with

dummy variables for 2nd, 3rd, and 4th year (with 1st

year as the reference).

The linear regression calculates a coefficient for

each predictor. Standard errors and p-values are

calculated for each coefficient to assess statistical

significance. The fit of the model was evaluated using

the coefficient of determination (R²), which indicates

the proportion of variance in sleep quality explained

by the predictors. The model predicts that if certain

factors (e.g., sleep duration or screen time) had a

strong linear relationship with sleep quality, their

coefficients would be significantly different from

zero and the R² would be notably above 0. This paper

expected strong predictors to have significant

coefficients (low p-values) and a high R².

To provide a comprehensive analysis, the

regression was performed on both the full dataset

(500 students) and a filtered dataset (52 students). The

filtered dataset included only students whose sleep

schedule data met strict validity criteria, allowing the

inclusion of two additional variables: Weekday Sleep

Duration and Weekend Sleep Duration. This

comparison aimed to investigate whether higher-

quality data would improve the model's explanatory

power.

3.3 Random Forest

Secondly, a random forest regression model is

applied. A random forest is an ensemble machine

learning method that builds many decision trees and

averages their predictions. Unlike linear regression, it

can capture complex relationships automatically,

Determinants of University Students’ Sleep Quality

633

including non-linear effects and interactions between

factors (Tan and Greenwood, 2021).

The same set of predictors is used for the random

forest model. To train and evaluate the model, this

paper splits the dataset into a training set and a test

set. 80% of the data (400 students) is used for training

the model and the remaining 20% (100 students) is

used for testing. The model was trained with 100

decision trees (n_estimators = 100) using default

parameters for depth and splitting criteria. The model

recorded the R² on the test set, as well as the root

mean squared error (RMSE) which gives an estimate

of the average prediction error in the same units as the

sleep quality scale. The feature importance scores

from the random forest are also extracted. These

scores indicate how much each predictor contributed

to reducing error in the model’s decision trees, giving

a ranking of which factors the model found most

useful for predicting sleep quality.

As with the linear regression, the random forest

was applied to both the full dataset and the filtered

dataset to compare the model’s performance across

different data quality levels. The filtered dataset

allowed the random forest to incorporate Weekday

Sleep Duration and Weekend Sleep Duration,

providing deeper insight into how sleep timing affects

sleep quality.

Comparing both methods allowed people to see

whether a more complex model like random forest

would improve prediction accuracy or like linear

regression.

4 RESULTS AND DISCUSSION

4.1 Linear Regression Results

The multiple linear regression model for the full

dataset (n = 500) showed very weak explanatory

power (Table 3). The R-squared was 0.011 and the

adjusted R-squared was -0.019. The F-test result was

F (15, 484) = 0.370, p = 0.986, which indicates that

the model was not statistically significant. All

predictors had p-values above 0.05. Among them,

Caffeine Intake (B = -0.011, p = 0.889) and Weekday

Sleep End (B = 0.055, p = 0.633) showed slightly

larger coefficients, but the results were still not

significant. This suggests that the full dataset did not

provide useful predictors of sleep quality.

The model using the filtered dataset (n = 52) had

a higher R-squared of 0.331 (Table 4), though the

adjusted R-squared was slightly negative (-0.003).

The F-test was F (17, 34) = 0.990, p = 0.491, which

was also not statistically significant. However, some

predictors showed stronger effects. Sleep Duration (B

= 0.406, p = 0.204) and Weekday Sleep Duration (B

= 0.454, p = 0.204) had the largest positive

coefficients. Caffeine Intake (B = -0.601, p = 0.073)

showed a negative effect and had the smallest p-

value, though it was still above 0.05.

Table 3: Linear Regression (n = 500)

Variables B P Value

Constant 4.869 0.019*

X4_4 0.115 0.771

X4_3 0.277 0.463

X4_2 0.110 0.773

X3_2 -0.110 0.750

X3_1 -0.479 0.140

X2 0.018 0.758

X5 -0.042 0.649

X6 0.050 0.207

X7 0.057 0.721

X8 -0.011 0.889

X9 -0.001 0.823

X10 -0.008 0.740

X12 -0.001 0.970

X11 0.055 0.633

X13 -0.020 0.869

R-squared 0.011

Adjust R-squared -0.019

F Test F (15, 484) = 0.370, p =

0.986

Table 4: Linear Regression (n = 52)

Variables B P value

Constant 4.222 0.575

X4_4 0.644 0.676

X4_3 0.188 0.890

X4_2 -0.359 0.805

X3_2 0.476 0.701

X3_1 -1.238 0.309

X2 -0.215 0.333

X5 0.406 0.204

X6 -0.206 0.187

X7 0.418 0.464

X8 -0.601 0.073

X9 -0.002 0.914

X10 -0.064 0.639

X12 0.098 0.617

X11 -0.310 0.584

X13 0.607 0.365

X14 0.454 0.204

X15 -0.172 0.704

R-squared 0.331

Adjust R-squared -0.003

F Test F (17, 34) = 0.990

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

634

When comparing the two models, the filtered dataset

provided clearer patterns and larger coefficients for

key variables like Sleep Duration and Caffeine Intake.

Although neither model reached statistical

significance, the filtered dataset showed improved

model fit and stronger trends. This suggests that

cleaning the data and focusing on valid observations

can help improve the performance of regression

models, even if the results are still limited by sample

size.

4.2 Random Forest Results

The random forest model with the full dataset (n =

500) also identified Study Hours (weight = 0.128),

Weekday Sleep Start (weight = 0.126), and Sleep

Duration (weight = 0.097) as the top contributors.

Other variables such as Physical Activity and

Weekend Sleep Start had moderate importance, but

most of the demographic variables, like University

Year and Gender, had very low weights, suggesting

minimal influence on sleep quality (Table 5 and 6).

Table 5: Feature Weight from Random Forest Model

Variables Weight

X4_4 0.011

X4_3 0.010

X4_2 0.010

X4_1 0.010

X3_2 0.008

X3_1 0.011

X3_3 0.013

X2 0.046

X5 0.097

X6 0.128

X7 0.074

X8 0.045

X9 0.092

X10 0.126

X12 0.110

X11 0.103

X13 0.107

Table 6: Model Evaluation (n=500)

Indicator Training

Set

Test

Set

R-squared 0.849 -0.058

RMSE 1.156 3.007

In addition to analyzing feature importance, the

performance of the random forest models was

assessed using R-squared and Root Mean Squared

Error (RMSE). For the full dataset (n = 500), the

model had an R-squared of 0.849 on the training set

but dropped to -0.058 on the test set, indicating

overfitting and poor generalization. The RMSE

values were 1.156 for the training set and 3.007 for

the test set, further confirming that the model did not

perform well on unseen data.

Table 7: Feature Weight from Random Forest Model

(n=52)

Variables Weight

X4_4 0.003

X4_3 0.015

X4_2 0.015

X4_1 0.004

X3_2 0.023

X3_1 0.014

X3_3 0.004

X2 0.027

X5 0.080

X6 0.134

X7 0.059

X8 0.109

X9 0.042

X10 0.069

X12 0.050

X11 0.043

X13 0.100

X14 0.138

X15 0.074

Table 8: Model Evaluation (n=52)

Indicator Training

Set

Test

Set

R-squared 0.832 0.054

RMSE 1.273 2.323

The random forest model using the filtered dataset (n

= 52) revealed that the most important predictors of

sleep quality were Study Hours (weight = 0.134),

Weekday Sleep Duration (weight = 0.138), Weekend

Sleep End (weight = 0.100), Sleep Duration (weight

= 0.080), and Weekday Sleep Start (weight = 0.069).

These variables contributed the most to the model’s

prediction accuracy. Notably, Weekday Sleep

Duration and Study Hours showed the highest

weights, indicating that both the length of sleep and

academic workload might have a relatively larger

influence on students’ sleep quality in the filtered

dataset (Table 7 and 8).

Determinants of University Students’ Sleep Quality

635

For the filtered dataset (n = 52), the random forest

model achieved an R-squared of 0.832 on the training

set and 0.054 on the test set. The RMSE values were

1.273 (training) and 2.323 (test). While the test set R-

squared was still low, it was slightly better than the

result from the full dataset, suggesting a marginal

improvement in predictive stability when using the

filtered data.

Comparing the two models, both results highlight

Study Hours and Sleep Duration as consistently

important factors. However, the filtered dataset

emphasized Weekday Sleep Duration and Weekend

Sleep End more strongly, while the full dataset placed

greater weight on Weekday Sleep Start. This suggests

that when outliers and unrealistic data were removed,

the model focused more on total sleep duration as a

key predictor, while the full dataset’s model was

slightly more sensitive to the timing of sleep.

4.3 Comparison with Previous Studies

Some researchers have also analyzed the same

dataset. For example, Tapendu used both multiple

linear regression and random forest models. In their

work, they added several new variables, including

Average Sleep Duration and Sleep Onset Time

Difference, which aimed to describe students’ sleep

habits in more detail. Their linear regression model

achieved a relatively high R-squared, and their

random forest results showed that Sleep Duration,

Average Sleep Duration, and Weekday Sleep End

were the most important factors influencing sleep

quality. These results are generally consistent with

the trend found in this study, where sleep duration-

related variables also appeared to be key predictors.

However, there are clear differences between their

approach and mine. This paper focused more on

improving data quality through strict filtering. The

filtered dataset contained only records with valid

sleep schedule data, which made it possible to include

two new variables: Weekday Sleep Duration and

Weekend Sleep Duration. Although my model

performance was lower, the focus was on ensuring

that only reliable data were used. These differences in

variable selection and data processing may explain

why the results are not exactly the same (Tapendu,

2024). Overall, both studies highlight that sleep

patterns play an important role in determining sleep

quality, but different methods can lead to different

outcomes.

5 CONCLUSION

This study explored how student lifestyle factors

affect sleep quality using both linear regression and

random forest models. The results showed that

neither model could accurately predict sleep quality,

with low R² scores on the test set for both models.

Although the random forest model performed well on

the training data, its performance dropped

significantly on the test set, suggesting overfitting.

Still, both methods consistently highlighted sleep

duration as one of the most important factors

influencing sleep quality.

There are also some limitations. The dataset did

not include psychological factors such as stress,

anxiety or depression, which may strongly influence

sleep, especially for international students who are

always influenced by culture shock. Environmental

factors like sunlight duration and seasonal changes

were also not considered.

Moreover, international students often face extra

challenges that can affect their sleep quality. In future

research, more detailed data should be collected

especially data about mental health, social factors.

With better and comprehensive data, models may

give more useful and accurate predictions of student

sleep quality.

REFERENCES

Becker, S. P., Jarrett, M. A., Luebbe, A. M., Garner, A. A.,

Burns, G. L., Kofler, M. J. 2018. Sleep in a large, multi-

university sample of college students: sleep problem

prevalence, sex differences, and mental health

correlates. Sleep health, 4(2), 174-181.

Hershner, S. D., Chervin, R. D. 2014. Causes and

consequences of sleepiness among college

students. Nature and science of sleep, 6, 73-84.

Lemma, S., Berhane, Y., Worku, A., Gelaye, B., Williams,

M. A. 2014. Good quality sleep is associated with better

academic performance among university students in

Ethiopia. Sleep & breathing = Schlaf & Atmung, 18(2),

257-263.

Lund, H. G., Reider, B. D., Whiting, A. B., Prichard, J. R.

2010. Sleep patterns and predictors of disturbed sleep

in a large population of college students. The Journal of

adolescent health: official publication of the Society for

Adolescent Medicine, 46(2), 124-132.

Schmickler, J. M., Blaschke, S., Robbins, R., Mess, F. 2023.

Determinants of Sleep Quality: A Cross-Sectional

Study in University Students. International journal of

environmental research and public health, 20(3).

Sohn, S. I., Kim, D. H., Lee, M. Y. Cho, Y. W. 2012. The

reliability and validity of the korean version of the

pittsburgh sleep quality index. Sleep & Breathing, 16(3),

803-812.

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

636

Tan, C. P. Y., Greenwood, K. M. 2021. Stress, Sleep and

Performance in International and Domestic University

Students. Journal of International Students, 12(1), 81-

100.

Tapendu. 2024. Student Sleep Pattern Analysis. Kaggle.

https://www.kaggle.com/code/iamtapendu/student-

sleep-pattern-analysis#Multiple-Regression-Analysis

Yang, P. Y., et al. 2012. Exercise training improves sleep

quality in middle-aged and older adults with sleep

problems: a systematic review. Journal of

Physiotherapy, 58(3), 157-163.

Zhang, L., Zheng, H., Yi, M., Zhang, Y., Cai, G., Li, C.,

Zhao, L. 2022. Prediction of sleep quality among

university students after analyzing lifestyles, sports

habits, and mental health. Frontiers in psychiatry, 13,

927619.

Determinants of University Students’ Sleep Quality

637