A Comparative Analysis of Contact and Non-Contact Approaches

Using Machine Learning for Gaming Disorder Detection

Zhaochen Jiang

College of Liberal Art & Science, University of Illinois Urbana-Champaign, Champaign, Illinois, 61820, U.S.A.

Keywords: Gaming Disorder, Machine Learning, Gaming Addiction Detection.

Abstract: With the rapid development of the Internet and the gaming industry, video games have become a major form

of entertainment, leading to an increase in gaming addiction, which was officially classified as a mental

disorder by the World Health Organization in 2018. This article first introduces the concept and current status

of gaming disorder and then reviews the application of machine learning (ML) in identifying gaming addiction,

analyzing existing research on contact, e.g., Electroencephalogram (EEG), Functional Near-Infrared

Spectroscopy (fNIRS), and non-contact, e.g., questionnaires, gaming data, methods. This review focuses on

various machine learning techniques, such as support vector machines, random forests, and deep learning

models, and their applications in improving the accuracy and efficiency of addiction diagnosis. The use of

ML to study physiological signals and behavioral indicators has achieved encouraging results, although there

are still limitations in the generality of the models and data acquisition methods. This article compares

different ML methods, explores their advantages and limitations, and proposes potential improvements for

future research on gaming disorder detection.

1 INTRODUCTION

With the continuous development of the Internet and

games, games are gradually becoming a major form

of entertainment for modern people. Compared with

many traditional forms of entertainment, such as

chess, sports, and performances, electronic games are

relatively low-cost, convenient, and do not require

venues. However, while the game industry is

booming, although it will not cause great harm to the

body, games can also cause people to have varying

degrees of addiction, just like gambling, alcohol, and

drugs. The World Health Organization included

gaming disorders in the category of mental and

behavioral disorders in 2018. Globally, the

prevalence of game addiction shows a trend that the

prevalence in Asia, especially in East Asia, China,

Japan, and South Korea, is higher than in other

regions, and the prevalence of adolescents can reach

15% (World, 2020). In terms of the proportion of

patients, there are more young people than the elderly

and more men than women. Game addiction will

affect the physical and mental health of patients to

varying degrees (Paulus, 2018). Physiological

damage includes impaired vision caused by long-term

use of the eyes and back problems caused by long-

term sitting.

Mental health problems include difficulty

sleeping and reduced self-control. Compared with

non-addicts, game addicts are more likely to

experience negative emotions such as depression,

anxiety, and loneliness, which affects academic

performance and quality of life (Dong, 2022).

Therefore, from parents to schools to the government,

the issue of adolescent gaming disorder is gradually

being taken seriously, and different laws and

regulations are being introduced to change this

phenomenon. The characteristics of gaming disorder

include decreased control over gaming, continuing to

play games even when negative effects have occurred

or are about to occur, and prioritizing gaming over

other activities (Dong, 2022). From a neurobiological

perspective, like other material or non-material

addictions, such as alcohol and gambling, the

distribution of cerebral blood flow in the brain of

game addicts is different from that of normal people,

and the reward mechanism and self-control area of

the brain are affected. Most of the current treatment

methods can be divided into drug therapy and other

types of treatment. Other types of treatment include

family care, psychological counseling, etc. Most drug

Jiang, Z.

A Comparative Analysis of Contact and Non-Contact Approaches Using Machine Learning for Gaming Disorder Detection.

DOI: 10.5220/0013332600004558

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

In Proceedings of the 1st International Conference on Modern Logistics and Supply Chain Management (MLSCM 2024), pages 387-391

ISBN: 978-989-758-738-2

387

therapy is used to relieve depression and anxiety to

reduce the impact of addiction. However, the issue of

how to treat gaming addiction has aroused heated

discussions among many scholars. Some scholars

believe that games are also a form of entertainment

for teenagers, just like other activities, such as sports,

chess cards, and comics. If the game time is long, it is

not enough to determine that the player is a game

addict, because if there is no game, he/she will choose

other activities as substitutes to meet the needs of

leisure and entertainment (Stevens, 2021). Therefore,

whether from a psychological or physiological

perspective, such as the self-rating scale for gaming

disorder or Electroencephalogram (EEG), it is very

important to accurately identify addicts and provide

effective treatment.

As a powerful data analysis tool, machine learning

has performed well in areas such as behavioral pattern

recognition and classification tasks. It can identify

potential addictive behavior patterns by analyzing

many user behavior data (Jordan, 2015). This method

can not only improve the efficiency of recognition but

also provide real-time, data-based diagnostic support.

If the original data of game players, such as EEG and

their self-rating scale, can be classified and identified

through machine learning models, the accuracy and

efficiency of recognition will be greatly improved

(Costa, 2019). Therefore, although it cannot provide

direct treatment, its advantages of high efficiency and

high accuracy have made machine learning gradually

applied to the field of game addiction identification.

Some studies have tried to apply various machine

learning algorithms, such as support vector machines

(SVM), random forests, deep learning, etc., to the

problem of identifying game addiction, and have

achieved remarkable results.

This paper aims to review existing related

research. First, the application of different machine

learning methods in identifying game addiction is

classified. Secondly, select several studies in each

application category and analyze their research

methods, model types, etc. Next, analyze the results

of each article and compare the advantages and

disadvantages of different models. Finally, based on

the author's information, supplement the deficiencies

of the method or model.

2 MACHINE LEARNING-BASED

DETECTION OF GAMING

DISORDER

2.1 Overview

At present, the research on the degree of addiction of

adolescents to online games using machine learning

is mainly divided into two types, namely contact and

non-contact. The contact type mostly uses the

relevant model of machine learning to analyze the

brain wave signal data of the subjects to determine

whether they are addicted to the Internet and the

severity of the degree. For non-contact game

addiction research, the main method is to use machine

learning models and algorithms to analyze the

questionnaire results filled out by the subjects or the

game-playing data of the subjects, to determine

whether they are addicted to games.

2.2 Contact Gaming Disorder

Detection

An existing contact machine learning game addiction

judgment method Functional Near-Infrared

Spectroscopy (fNIRS), namely functional Near-

Infrared Spectroscopy (Bunce, 2006). According to

the research of Wang Qiwen and others, they used

this technology to measure the changes in the

prefrontal cerebral blood oxygen concentration when

the subjects' brains performed the stop signal task.

Then machine learning to analyze and compare with

the data of normal people to determine whether the

subject is a game addict. The experimental method is

to let the subjects use the left and right arrow keys on

the keyboard to respond to the left and right arrows

on the screen according to the requirements under

interference-free conditions and use the light NIRS

portable brain imaging device worn on the subject's

head to use near-infrared light to monitor the changes

in blood oxygen concentration in the forehead of the

brain in real-time. After collecting the raw data,

feature extraction was performed on it. In this study,

the mean, skewness, and kurtosis were mainly

extracted. Afterward, the researchers selected three

machine learning classifiers and a Long Short-Term

Memory (LSTM) classification model to distinguish

between game addicts and healthy people. The three

machine learning classifiers are SVM, Linear

Discriminant Analysis (LDA), and K-Nearest

Neighbors (KNN). In this experiment, their

classification accuracy rates of the combined mean,

skewness, and kurtosis were 67.4%, 63.6%, and

MLSCM 2024 - International Conference on Modern Logistics and Supply Chain Management

388

71.2%, respectively. The accuracy of the LSTM

model in identifying game addicts was 85.7, which is

better than the traditional machine learning method.

This study shows that the combination of fNIRS

signals and machine learning classification can

provide powerful clinical treatment assistance for

identifying whether patients are addicted to games

(Cho, 2022).

Another example of a contact gaming disorder

determination model was conducted by Lee et al. The

original data was obtained through the EEG of the

subjects in the eyes closed and eyes open state for five

minutes each. After preprocessing the initial EEG

signal, it was divided into multiple periods and

converted into corresponding frequency bands. In

terms of algorithm selection, the researchers selected

three EEG-related algorithms. The researchers used

cross-validation and took into account the weighted

calculation of the model's bias and split probability.

After multiple iterations and repetitions, the model's

ability to correctly identify whether a subject is

addicted to games gradually became evaluable. The

results showed that compared with the unimodal

model, the multimodal model is more accurate in

identifying online gaming disorders. However, the

researchers said that this study still has the problem

of small research subjects and lack of universality of

the results. And further combining deep learning

algorithms and machine learning is also necessary to

improve model performance (Lee, 2022).

2.3 Non-Contact Gaming Disorder

Detection

There are two main types of non-contact game

addiction judgment models. One is to judge by

analyzing the game-playing data of students or

players, such as calculating from the game time,

campus network records, and student attendance. The

other is to let the subjects fill out a questionnaire

about game addiction using an evaluation form and

judge whether the subjects are game addicts by

analyzing the scores of each item.

Song's research belongs to the first category. In

this model, the evaluation indicators of whether it is

Internet addiction are divided into two categories:

impact and daily behavior. Impact includes three

subcategories: damage to physical and mental health,

frustration of willpower, and failure to graduate

normally. Daily behavior consists of five

subcategories: the degree of attention parents pays to

education, academic performance, average daily

Internet time in the dormitory, class attendance, and

mental health. In the model for judging Internet game

disorder, each data has different data characteristics,

which will greatly affect the results. The researchers

used fuzzy C-means clustering to divide the data. In

the calculation and evaluation stage of the data, the

researchers selected the fuzzy hierarchical analysis

method for quantitative and qualitative analysis and

divided the degree of Internet game addiction of

specific students into four levels, namely normal,

mild, severe, and extremely severe. When assigning

weights to each indicator, the researchers took into

account the theory that multiple factors in psychology

can affect people's judgments and the method theory

of multi-attribute decision-making. The model uses

consistency tests to perform data fuzzification and

weight calculation. The fuzzy operator uses the main

factor-determined Zade operator. The researchers

pre-set several corresponding decisions for different

game addiction severity ratings, such as

psychological counseling, contacting parents, and

pushing reminders. The above measures can be

changed according to actual conditions. In addition,

the researchers also distributed questionnaires in

colleges and universities as a control for this model

and classified the data filled out by each respondent

into data applicable to this model. In this study, the

researchers gave a more detailed data fuzzy

processing calculation method, first using the Fuzzy

C-Means clustering method to divide the data, and

then using the fuzzy hierarchical analysis method to

determine the severity of addiction. However, this

study did not give a detailed questionnaire survey

result for the model's assistance and machine learning

process. At the same time, for some raw data that are

difficult to quantitatively analyze, such as the

importance of parental education, and raw data that

are difficult to obtain, such as mental health, the

researchers did not give detailed acquisition methods

and quantitative calculation methods (Song, 2019).

The study by Kong et al. belongs to the second

non-contact measurement model. They selected

2,100 students from three junior high schools and

three high schools in three regions of Guizhou

Province as the research subjects. The questionnaire

design includes Nine-item Internet Gaming Disorder

Scale-Short Form (IGDS9-SF), Parental

Psychological Control and Autonomy Support

Questionnaire (PPCASQ), Motivational Structure

Questionnaire, Relative Deprivation Questionnaire,

Deviant Peer Interaction Questionnaire and Self-

Control Dual System Scale for data collection. Except

for the relative deprivation questionnaire, all other

questionnaires are scored using the Likert "1 to 5

points" system, and the relative deprivation

questionnaire uses the Likert "1 to 6 points" system.

A Comparative Analysis of Contact and Non-Contact Approaches Using Machine Learning for Gaming Disorder Detection

389

When the research subjects fill in the questionnaire,

there are professionals to explain, so that the subjects

can give more credible data with full understanding.

And the diagnostic accuracy of IGDS9-SF reaches

96.1%. Based on Python, the researchers use

algorithms including logistic regression, support

vector machine, decision tree, gradient boosting tree,

adaptive boosting algorithm, and random forest to

judge and predict game addiction behavior.

According to the research results, the prediction

accuracy of these six methods is higher than 90%.

After comprehensively considering the evaluation

indicators such as precision and recall rate, the

adaptive improvement model performs best. The

adaptive improvement model continuously iterates to

enhance the weight ratio of each item and focuses on

the distribution pattern of the data with incorrect

predictions to complete the correction. In this study,

the prediction accuracy reached 99%. However, since

only six schools in three places in Guizhou were

selected in this study, the geographical scope is small,

and its model may need more data training to improve

universality. In addition, the method of obtaining raw

data through questionnaires is more subjective than

that of obtaining data from student behavior and

players, so the data may be biased (Kong, 2024).

3 DISCUSSIONS

Although some machine models in some of the

studies mentioned above have achieved high

accuracy in detecting gaming disorders, each study

may have problems to a greater or lesser extent.

Although the detection accuracy of both in specific

model algorithms has reached a high level, that is,

more than 80%. However, in comparison, the

accuracy of the non-contact model is higher than that

of the contact model. This may be because the EEG

data is more complicated for all Likert scales, and

there are more misjudgements and data losses in the

judgment process. However, the latter needs to design

weights for various indicators, and there is a problem

that a lot of statistical calculation modeling is

required. The problem of non-contact machine model

detection is more serious. For the questionnaire of the

non-contact detection model, since the subjects

themselves conduct self-evaluation in the

questionnaire survey, for example, the scales of

parental family control, self-esteem, and anxiety are

difficult to quantify and there are subjective

misjudgements. Therefore, this will more or less

affect the original data submitted to the model for

processing, thereby affecting the accuracy of the

judgment. In general, both non-contact and contact

models have the problem of lack of universality. Lee

et al.'s EEG model study has few research subjects, so

it lacks universality. Although Song's fuzzy multi-

attribute college student Internet use disorder

prevention and treatment platform gives a relatively

detailed description of the model research and

statistics, according to the author's point of view, the

model is only a proposed concept. If people want to

better apply this model to detect and prevent college

students' game addiction, the support of the school is

required. In addition, some of the data used in the

study, such as Internet usage time and student

attendance, may be misjudged, require various data

associations within the school, and have privacy

issues. More research and investigation are needed

before it can be put into practical use. However, the

game addiction prevention and treatment methods in

this study are more than other studies, and are not the

most commonly used psychological cognitive

therapy for game addiction treatment, which has

certain reference value. For the study by Kong et al.,

they selected students from six high schools in three

regions of Guizhou Province for research. Although

the results are relatively accurate, there will be more

serious regional and age biases. For example, data

from eastern and northern China, as well as college

students, are not covered. However, the results given

by their machine model are only relatively accurate

for the existing research subjects, and the accuracy of

the different models used is more than 90%. In

addition, most of these studies use the game use

disorder evaluation scale as a control to determine

whether the machine is successful. However, as

mentioned above, although the Likert scoring method

in the scale assists in quantifying the scores of various

data, its authenticity still needs to be investigated. But

overall, most of the above machine learning models

perform well. Although there are some problems, the

accuracy of the model itself has been verified in the

study. If there are conditions to conduct more

extensive research in the future, or if there are other

more accurate methods to identify game addicts as

control and training data, then using machine learning

to assist in identifying gaming disorders will greatly

improve the efficiency and accuracy of treatment. If

this technology can overcome the problems of control

data and universality, then its high efficiency and high

accuracy have the potential and prospects for large-

scale use.

MLSCM 2024 - International Conference on Modern Logistics and Supply Chain Management

390

4 CONCLUSIONS

The accuracy of the machine learning models for

detecting gaming disorders reviewed and analyzed

above is relatively high, and they can complete

basically accurate detection under experimental

conditions. However, there is still room for

improvement in the experimental process, and there

is still a lack of more universal optimization before a

large number of practical applications. At the same

time, the formulation of various game time evaluation

scales should also take into account practical

problems such as supervisor bias. In addition, the

concept of game addiction is constantly changing

with the development of society and the game

industry. Not everyone who plays games for too long

means losing control of the game, such as e-sports

practitioners. In actual applications, in order to better

solve the problem of gaming disorders in society,

especially for young people, in addition to complete

and efficient detection methods, there should also be

a good social environment and appropriate treatment

methods. Current cognitive therapy can reduce

dependence on games, and drug treatment can reduce

anxiety and depression. However, in order to

fundamentally solve the problem of gaming disorders,

it requires the attention of society, schools and

families. If there are complete laws as protection and

abundant entertainment and sports resources as a

substitute, then gaming addiction will be a

fundamental treatment.

REFERENCES

Bunce, S. C., Izzetoglu, M., Izzetoglu, K., Onaral, B., &

Pourrezaei, K. 2006. Functional near-infrared

spectroscopy. IEEE engineering in medicine and

biology magazine, 25(4), 54-62.

Cho, T. H., Nah, Y., Park, S. H., & Han, S. 2022. Prefrontal

cortical activation in Internet Gaming Disorder Scale

high scorers during actual real-time internet gaming: A

preliminary study using fNIRS. Journal of Behavioral

Addictions, 492-505.

Costa, S., & Kuss, D. J. 2019. Current diagnostic

procedures and interventions for gaming disorders: A

systematic review. Frontiers in Psychology, 10, 578.

Dong, X., Jiang, Y., Zhang, Y., Zhang, W., Wang, N. 202

2. Hazards of game addiction to health in adolescents.

Shanghai Journal of Preventive Medicine, 34(5): 504-

508.

Jordan, M. I., & Mitchell, T. M. 2015. Machine learning:

Trends, perspectives, and prospects. Science, 349

(6245), 255-260.

Kong, W., Wang, K., Tuo, A., Li, B., Zheng, Q., Jiang, H.

2024. Building a predictive model for adolescent

Internet gaming disorder based on machine learning.

Chinese Journal of School Health, 45(8): 1080-1085

Lee, J. Y., Choi, C. H., Park, M., Park, S., & Choi, J. S.

2022. Enhanced resting-state EEG source functional

connectivity within the default mode and reward-

salience networks in internet gaming

disorder. Psychological medicine, 52(11), 2189-2197.

Paulus, F. W., Ohmann, S., Von Gontard, A., & Popow, C.

2018. Internet gaming disorder in children and

adolescents: a systematic review. Developmental

Medicine & Child Neurology, 60(7), 645-659.

Song, H. 2019. Research on evaluation and control model

of undergraduates’ online games addiction based on

Fuzzy Multiple Attribute Decision making under the

background of smart campus. Journal of Chongqing

University of Technology (Social Science), 7, 89-96

Stevens, M. W., Dorstyn, D., Delfabbro, P. H., & King, D.

L. 2021. Global prevalence of gaming disorder: A

systematic review and meta-analysis. Australian & New

Zealand Journal of Psychiatry, 55(6), 553-568.

World Health Organization, 2020, Addictive behaviours: G

aming disorder. URL: https://www.who.int/news-room

/questions-and-answers/item/addictive-behaviours-ga

ming-disorder. Last Accessed: 2024/09/20.

A Comparative Analysis of Contact and Non-Contact Approaches Using Machine Learning for Gaming Disorder Detection

391