Abstract:
While smartphone addiction is becoming a recent concern with the exponential increase in the number of smartphone users, it is difficult to predict problematic smartphone users based on the usage characteristics of individual smartphone users. This study aimed to explore the possibility of predicting smartphone addiction level with mobile phone log data. By Korea Internet and Security Agency (KISA), 29,712 respondents completed the Smartphone Addiction Scale developed in 2017. Integrating basic personal characteristics and smartphone usage information, the data were analyzed using machine learning techniques (decision tree, random forest, and Xgboost) in addition to hypothesis tests. In total, 27 variables were employed to predict smartphone addiction and the accuracy rate was the highest for the linear support vector machine model and the lowest for the decision tree model . The results showed that users’ general information, such as age group, job classification, and sex did not contribute much to predicting their smartphone addiction level. The study can provide directions for future work on the detection of smartphone addiction with log-data, which suggests that more detailed smartphone’s log-data will enable more accurate results.

Keywords: dataset of smartphone , machine learning algorithm decision tree ,linear support Vector machine

Introduction:

With the rapid increase in smartphone penetration, they are becoming a part of our daily lives. Due to the various functions and convenience of smartphones, the number of users worldwide was more than 1.08 billion in early 2012, and continues to increase exponentially . In the UK, 68% of adults were reported to own a smartphone, and the number of smartphone users in South Korea exceeded 39 million in early 2012 . This trend is observed worldwide . On the other hand, the number of people who depend too much on smartphones is also increasing. According to a survey of 29,712 smartphone users conducted by Korea Internet and Security Agency (KISA) in 2017, the high-risk smartphone addiction rate was about 18.6% (7860), which increased by about 1% since the previous year . Further, the proportion of high-risk smartphone usage groups by age in the past two years increased from 6.7% to 19.1% among infants, the biggest increase among all age groups, followed by adults, with an increase from 3.9% to 17.4%. Among young adults and college students, the increasing reliance on smartphones has created a new potential for the widespread abuse of the technology in ways that suggest addiction. Many researchers call attention to the harmful effects of smartphone overuse, and various studies have been conducted . While various studies have focused on smartphone addiction, there is little research on the prediction of smartphone addiction based on usage pattern. Thus, the present study aimed to identify factors that predict the smartphone addiction level, such as smartphone usage level by contents, gender, age, and job. Accordingly, it attempted to develop a model to predict smartphone addiction level. For prediction, machine learning methods that are used commonly were tested. Although smartphone addiction does not easily fit into the standard classification of impulse disorders according to the Diagnostic and Statistical Manual of Mental Disorders (DSM-5), the concept of smartphone addiction is increasingly becoming accepted . The following problematic behaviors associated with smartphone use are most suggestive of addiction: (1) use in dangerous situations; (2) harm or repeated interruptions to work, social life, family life, and/or physical and mental well-being; (3) urges and drives to repeat behavior; (4) dependence, tolerance, and increasing need for stimulation to achieve satisfaction; and (5) anxiety or negative feelings associated with inability to send or receive immediate responses . Similar symptoms have been reported in , and these have led some research to classify problematic smartphone use as a potential behavioral addiction . Excessive use of smartphones can interfere with everyday life and can cause various physical, mental, and social problems . A variety of activities can be conducted through one device, and most of the activities are classified into sedentary behavior, which is characterized by an energy expenditure of less than 1.5 metabolic equivalents
(METs). Such behavior results in low levels of energy expenditure and correlates with health problems such as obesity or metabolic syndrome . Excessive smartphone users showed less physical activity, such as a smaller number of steps taken per day, and they tended to consume fewer calories per day . Other adverse physical effects include neck pain symptoms , craniocervical posture, dry eyes, carpal tunnel syndrome, sleep disturbance, and headaches.

Regarding mental health, smartphone overuse might be related to depression and anxiety . Additionally, it can cause human relationship problems and reduce academic achievement . A report published by KISA stated that “45.8% of smartphone users feel anxiety when they are not holding their smartphone, 27.1% spend too much time using their smartphone, and 22.6% have repeatedly attempted to reduce their smartphone use but have always failed. Moreover, 21% of smartphone users reported difficulties with school or work due to excessive smartphone use” . Thus, smartphone use can become a serious problem because it involves the use of several addictive elements such as the internet and games . Considering the importance of problems caused by excessive smartphone use, it is important for users to be aware of their condition. In South Korea, the Smartphone Addiction Scale (S-scale) was developed by KISA to assess the current level of addiction . The questionnaire consists of 10 questions to identify the level of smartphone addiction risk and to distinguish the high-risk group. While it provides useful information about the user’s current level of addiction, the questionnaire does not provide information on smartphone usage patterns, since it aims to diagnose the user’s psychological factors. Thus, even if classified as over-dependent by the S-scale, information about usage patterns cannot be obtained, and practical guidelines on addressing addiction problems cannot be developed. To identify the predictors of smartphone addiction, the data should be analyzed with multiple variables in an integrated manner.

Although some mobile applications capture log data from smartphones, they have not been
expanded to determine whether users are over-dependent . Our research is motivated by the lack of results and our goal is to identify the predictors of smartphone addiction. In this paper, we propose a model for predicting age-based smartphone addiction level using information available from smartphones, by exploring user groups that exhibit similar usage patterns. Our research is based on a survey called “A Survey on Smartphone Over-Reliance (SSOR)”. The survey is conducted by KISA every year and it comprises about 180 questions. The data are collected from 29,712 participants (14,790 male and 14,922 female) in 2017. The survey and the results of the respondents’ Smartphone addiction scale survey were integrated and analyzed. While our study relies on survey data (not actual usage information), a sufficient sample size was obtained. First, we conducted hypothesis tests to find out whether there is a difference in smartphone usage patterns and monthly spending between the normal user group and the problematic user group. The goal of the hypothesis test is to test whether there is a difference in the content used by the two groups of users. Except for the use of office search, web document and sports betting, there were significant differences in the usage level of all contents. In addition, significant differences were observed in the average monthly expenditure on games, movies/videos, and e-books and cartoons between two groups. In particular, those in the problematic group spent 2, 1.64, and 2.4 times more money on games, movies/videos, and e-books and cartoons, respectively, as compared to those in the normal group. As such, through the t-test, it is possible to figure out how different the usage patterns and expenditures are between the two groups. In addition to hypothesis testing, machine learning techniques were also used to reveal the relationship between usage pattern and smartphone addiction level. The samples were randomly divided into a training set consisting of 70% of all observations and a test set consisting of the remaining 30% of observations. Three learning methods were considered and tested: random forest, Xgboost and decision tree. The average values for accuracy were 82.59% (random forest), 80.77% (Xgboost) and 74.56% (decision tree), respectively.

The following can be derived from our results. First, the results show that information on users’ smartphone usage patterns and expenditure can be used as predictors to determine whether users are addicted to smartphones. Existing questionnaires developed to determine smartphone addiction are mostly focused on the psychological factors of users. Because information on usage patterns provides more objective information, the results of this study can be used to improve existing surveys. Second, the results may serve as evidence of high complexity in the smartphone addiction diagnosis. In our model based on learning techniques, the user’s personal information, content usage patterns and expenditure are required to ensure the sufficient accuracy. This means that various factors (age or content use) together influence the determination of smartphone addiction. Third, our results provide a basis for developing programs such as self-diagnostic applications to detect smartphone addiction. Note that our model uses only the information stored in the smartphone as predictors. If we can diagnose the smartphone addiction level just from the information stored in a smartphone (without psychological factors), this can provide useful information to users through the form of applications or programs. While the limitation of our research is that we rely on survey data (not actual usage information). We hope that the study can provide directions for future work on the detection of smartphone addiction with inputs, which suggests that more detailed smartphone’s log-data will enable more accurate results.

Problem statement:

smartphone addiction is becoming a recent concern with the exponential increase in the number of smartphone users, it is difficult to predict problematic smartphone users based on the usage characteristics of individual smartphone users.

Objective:

This study aimed to explore the possibility of predicting smartphone addiction level with mobile phone log data. By Korea Internet and Security Agency (KISA), 29,712 respondents completed the Smartphone Addiction Scale developed in 2017. Integrating basic personal characteristics and smartphone usage information, the data were analyzed using machine learning techniques (decision tree, random forest, and Xgboost) in addition to hypothesis tests.