Social Media Addiction in Adolescents and the Effect of Emotional Intelligence on Ghosting Authors: Ramazan İnci, Sevgi Köroğlu Gökbel, Hatice Kiliçkiran Doğan, Nursel Gül, Kübra Yesin, Metin Yıldız Published: 2026 ABSTRACT Objective This study was conducted to examine the effect of social media addiction and emotional intelligence on ghosting among adolescents. Methods This descriptive cross-sectional study was conducted between June and September 2025 using an online survey method. The study was conducted with 425 adolescents aged 10–19 years living in Batman, Türkiye. R programming language 4.1.3, G*Power 3.1, and SPSS-22 programmes were used in the analysis of the study. Results It has been determined that the model is meaningful and usable in hierarchical regression analysis (F(2,422)=30.392, p=0.001). The level of social media addiction and emotional intelligence levels together explain 12.2% of the total variance in ghosting levels (R2=0.122). It was determined that an increase in participants’ social media addiction levels (t=4.895, p<0.001) and a decrease in their emotional intelligence levels (t=-4.021, p<0.001) were significantly associated with higher levels of ghosting. SHAP values show the extent to which each variable influences the model’s predictions and the magnitude of this influence. According SHAP values, social media addiction is identified as the most important variable in predicting the ghosting variable. Conclusion Participants’ increased social media addiction and decreased emotional intelligence were found to be associated with higher levels of ghosting. Longitudinal studies on ghosting among adolescents are recommended. Furthermore, the implementation of educational and intervention programs aimed at regulating social media use and enhancing emotional intelligence skills in adolescents is suggested. INTRODUCTION Social media addiction has become a widespread problem, particularly among adolescents. This addiction has been associated with individuals’ emotional intelligence. The ability to recognize, understand, and manage one’s own and other people’s emotions is known as emotional intelligence. Studies demonstrate that those with elevated emotional intelligence have a reduced susceptibility to social media addiction. For eISSN 1976-3026 OPEN ACCESS example, in Chen’s study [1], among college students who went through the COVID-19 pandemic, researchers discovered that those with higher levels of emotional intelligence had a lesser chance of being addicted to their smartphones. Furthermore, Jarrar and colleagues [2] examined the consequence of emotional intelligence on social media addiction and noted that high emotional intelligence may reduce use of social media sites. Adolescents’ ability to control their emotions has strong ties to the development of social media addiction. When studying the addictions of adolescents to gaming and social media, Giordano et al. [3] discovered that the ability to control one’s emotions is important. It has been noted that adolescents with deficient emotional management abilities are more prone to use rewarding behaviours such as social media to manage their emotional states. Kocabıyık and Bacıoğlu’s [4] research also shows that young people who use emotion-focused coping strategies are more prone to social media addiction. This situation causes adolescents to view social media use as an escape route. Ghosting is another important issue related to social media addiction. Ghosting refers to one person abruptly cutting off communication with another without explanation. Unlike social media addiction, which refers to problematic social media use, ghosting is a more specific relationship termination behavior that occurs within the context of interpersonal relationships. Despite this distinction, ghosting has become increasingly visible with the widespread use of digital communication environments, and research on the phenomenon has steadily increased [5,6]. When this occurs via social media, the role of emotional intelligence becomes more pronounced. Emotional intelligence encompasses an individual’s ability to perceive, understand, regulate, and effectively use both their own emotions and those of others in interpersonal relationships, which may be associated with behaviors such as ghosting [7,8]. Accordingly, individuals who experience difficulties in regulating their emotions and managing interpersonal communication may be more likely to engage in ghosting behavior during the process of relationship termination [9,10]. Similarly, it can be argued that individuals with lower levels of empathy and emotional awareness may be more likely to engage in ghosting, as they may be less likely to consider the negative emotional consequences that this behavior can have on the other person [9]. Particularly among adolescents, lower emotional intelligence may be associated with difficulties in social relationships and higher levels of ghosting [11]. The effects of emotional intelligence and social media addiction on adolescent ghosting behavior will be examined in the analysis. Although numerous studies have examined social media addiction, to the best of our knowledge, there is a limited number of studies investigating the role of social media addiction and emotional intelligence in ghosting among adolescents. Our findings will strengthen existing literature and provide new avenues for investigation. Research hypotheses H01: There is no significant relationship between the level of social media addiction and the level of ghosting among adolescents. H1: Higher levels of social media addiction are associated with higher levels of ghosting. H02: There is no significant relationship between the level of emotional intelligence and the level of ghosting among adolescents. H2: Lower levels of emotional intelligence are associated with higher levels of ghosting. METHODS The study was conducted using a descriptive cross-sectional research design. Data were collected between June and September 2025 through an online survey administered via Google Forms. The survey link was distributed through school groups and shared on WhatsApp and Instagram platforms. Participants were allowed to access the survey after being informed about the study and after the necessary parental/legal guardian consent had been obtained. Participants were also able to forward the survey link to other eligible individuals. Since all questions in Google Forms were set as mandatory, no missing data occurred. Due to the use of open online distribution and snowball sampling, the total number of individuals invited to participate and the response rate could not be determined. The study population consisted of adolescents aged 10–19 years living in Batman, Türkiye. The sample size was calculated using the formula n=(t2pq)/d2, which is commonly employed when the population size is unknown [12]. In the calculation, a 95% confidence level (t=1.96), p=0.50, q=0.50, and a sampling error of d=0.05 were assumed, resulting in a minimum required sample size of 384. At the end of the data collection process, 425 adolescents were reached, and the study was completed with 425 participants. After the study, researchers ran a post hoc power analysis on the 425 participants’ data and found that, with a medium effect size and 95% confidence level, the power was 99% [13]. The reporting of this study adhered to the STROBE standard [14]. Inclusion criteria All adolescents aged 10–19 who were able to communicate (without mental impairment or visual/hearing impairment), read and understand Turkish, and individuals that volunteered to participate were included into the research. Exclusion criteria Participants who could not engage in the assessment or communication due to certain circumstances, who were not aged between 10 and 19, or who were unwilling individuals who did not engage who were not included in the study. Data collection tools Following a literature review, the researchers developed the “Personal Information Form,” “Social Media Addiction Scale for Adolescents,” “Emotional Intelligence Scale for Adolescents,” and “Ghosting Scale” as instruments for data collecting. Personal Information Form This questionnaire, created by the researchers, includes inquiries about age, gender, educational attainment, income level, length of social media use, and other relevant factors. Social Media Addiction Scale for Adolescents The Adolescent Social Media Addiction Scale (ASMAS) was developed by Özgenel et al. [15] based on the Diagnostic and Statistical Manual of Mental Disorders-Fifth Edition criteria. The scale consists of 9 items rated on a 5-point Likert scale ranging from 1) never, 2) rarely, 3) sometimes, 4) often, to 5) always. The scale does not contain any reverse-scored items. Total scores range from 9 to 45, with higher scores indicating higher levels of social media addiction. Based on validity and reliability analyses, the developers reported that the scale is a valid and reliable measurement instrument. In the present study, the Cronbach’s alpha coefficient of the scale was calculated as 0.85. Emotional Intelligence Scale for Adolescents The Emotional Intelligence Scale, originally developed in South Korea by Lee and Kwak [16] and later adapted into Turkish by Kayıhan and Arslan [17], was used to assess participants’ perceptions of emotional competence. The scale consists of 20 items and three subdimensions. Items are rated on a 5-point Likert scale ranging from 1) strongly disagree to 5) strongly agree. Total scores range from 20 to 100, with higher scores indicating higher levels of emotional intelligence. To evaluate the psychometric properties of the scale, studies on internal consistency, item analysis, and factor analysis were conducted. Reliability analyses yielded a Cronbach’s alpha coefficient of 0.83 for the overall scale. In the present study, the Cronbach’s alpha coefficient was found to be 0.86. Ghosting Scale The Ghosting Scale, developed by Jahrami et al. [18], has a unidimensional structure and consists of 8 items. The items are rated using a 5-point Likert scale ranging from 1) never to 5) always. Total scores range from 8 to 40, with higher scores indicating higher levels of ghosting. The Turkish adaptation of the scale was conducted by Özalp et al. [19] Confirmatory factor analysis supported the original 8-item single-factor structure, and the model fit indices were found to be at acceptable levels. In addition, the scale demonstrated high internal consistency, with a reported Cronbach’s alpha coefficient of 0.88. In the present study, the Cronbach’s alpha coefficient of the scale was found to be 0.81. Ethical considerations Approval has been obtained from the Scientific Research and Publication Ethics Committee of Sakarya University to conduct the study (Date: 21.05.2025, Number: E-61923333-050.99-479838). The first page of the online survey contained an information sheet and an informed consent form. Participants were able to access the survey only after indicating their electronic consent. For participants under the age of 18, parental/legal guardian consent was obtained, whereas participants aged 18 years and older provided electronic informed consent themselves. Only data from individuals who provided consent were included in the study. The research was conducted anonymously, and no personally identifiable information, such as names, identification numbers, or contact details, was collected. All data were used solely for research purposes and were protected in accordance with confidentiality principles. The researchers adhered to the rules outlined in the Helsinki Declaration throughout the study. Data analysis The examination of the evaluation of data was accumulated utilizing SPSS 22.0 and G*Power 3.1 statistical applications. Percentages, arithmetic means, standard deviations, and minimum and maximum values were determined utilizing SPSS 22.0 (IBM Corp.). Essential normality tests were conducted throughout the data analysis, revealing that the data exhibited a normal distribution (kurtosis and skewness between -1.5 and +1.5) [20]. For statistical purposes, a p value lower than 0.05 was considered significant. Predictive analyses were conducted using R programming language version 4.1.3. To develop and evaluate the machine learning model, the dataset was randomly divided into training and testing subsets using a 70:30 ratio. The training set (70%) was used to train and optimize the model, whereas the testing set (30%) was reserved for independent performance evaluation. This approach was adopted to assess the generalizability of the model and reduce the risk of overfitting. Models established for predicting ghosting using K-Nearest Neighbors regression (KNN), Support Vector Machine regression (SVM), Artificial Neural Network regression (NN), Model Averaged Neural Network regression (avNNet), Random Forest regression (RF), XGBoost regression (XGBoost), Regression (REG [Elastic-net, Lasso, Ridge]), Monotone Multi-Layer Perceptron Neural Network regression (MONMLP), Quantile Regression Neural Network regression (QRNN), Decision Tree regression (DT), Principal Component Regression (PCR), and Gradient Boosting Machine regression (GBM) algorithms, and the determination of the best hyperparameters based on the logarithm-transformed training data. For each algorithm, hyperparameter optimization was performed to maximize the overall performance of the model. This process includes scaling the dataset between 0 and 1 and using a 10-fold cross-validation method. Data scaling or normalization ensures consistency among variables with different scales and allows algorithms to converge faster and more effectively. Cross-validation tests the model’s performance on unseen data, minimizing overfitting. This method randomly splits the dataset into 10 equal parts. In each iteration, a different part is used as the test set while the remaining 9 parts are used to train the model. This process is repeated until each part is used once as the test set. The results are calculated by averaging the performance metrics obtained across all iterations. This method ensures the model’s consistency and reliability across different datasets, preventing it from overfitting to the specific characteristics of the training set. Hyperparameter tuning was performed on the training dataset to ensure optimal performance of the algorithms. The results obtained from this process are the hyperparameter outcomes that emerged during the training phase after scaling the dataset between 0 and 1. These findings allowed the identification of optimal hyperparameters by comparing root mean square error (RMSE) values. Finally, the predictions obtained with the method that showed the best performance on both training and test data are compared with the test data. These visualizations detail the effectiveness of the methods used and the prediction success on the dataset. The most successful hyperparameter values are determined, and predictions are made on the test data. Upon examining the RMSE and mean absolute error (MAE) values, the test prediction results obtained using these optimal hyperparameters are determined. The RMSE and MAE values resulting from predictions made on the test data using the hyperparameters that yielded the most suitable metric values on the training dataset are presented. Although the analysis shows that all models except PCR, REG, and SVM produced similar and successful results, it was determined that the REG method provided the most accurate results. When the optimal hyperparameter values are used for the REG model, predictions are made on the test data. It visually presents the model’s predictions on the test data and how close these predictions are to the actual values. RESULTS It was established that 52.7% of the people involved in the research were female, 58.8% were attending high school, 54.1% had an income commensurate with their expenditures, 25.6% used social media for 2 hours per day, and their average age was 15.63±2.48 (Table 1). When examining the results of the hierarchical regression models developed to assess the relationship of social media addiction and emotional intelligence with the phenomenon of ghosting, the 95% confidence intervals for the B coefficients are presented in Table 2. The Model 1 was shown to be both significant and applicable according to the statistical assessments (F(1,423)=43.071, p=0.001). Social media addiction accounts for 9.0% of the overall variation in ghosting behavior (R²=0.090). According to the regression analysis, higher social media addiction scores were positively and significantly associated with higher ghosting scores (t=6.563, p<0.001) (Table 2). Model 2 statistical estimations point to the model’s significance and application (F(2,422)=30.392, p=0.001). Social media addiction and emotional intelligence together account for 12.2% of the overall variation in ghosting behavior (R²=0.122). The results of the regression model indicated that higher social media addiction scores (t=4.895, p<0.001) were significantly associated with higher ghosting scores, whereas lower emotional intelligence scores (t=-4.021, p<0.001) were significantly associated with higher ghosting scores (Table 2). Prediction using machine learning approach This research utilizes a dataset centered on ghosting. Hyperparameter tweaking was conducted on the training dataset to guarantee the algorithms’ best performance (Figure 1). The most successful hyperparameter values obtained from Figure 1 are determined, and predictions are made on the test data. The values (RMSE and MAE) shown in Figure 2 provide the test prediction results for these optimal hyperparameters. Figure 3 illustrates the model’s predictive performance on the test data. The SHAP values reveal the relative significance of each variable to identify the model’s predictions and the magnitude of this influence. The graph indicates that social media addiction is the primary variable in forecasting the ghosting variable (Figure 4). Figure 5 shows a three-dimensional surface plot of social media addiction and emotional intelligence as it relates to ghosting. All three of these variables social media addiction, emotional intelligence, and ghosting are shown in Figure 6, along with their distribution, interaction, and density graph (Figure 6). DISCUSSION The aim of this study was to examine the relationship between ghosting behavior, emotional intelligence, and social media addiction among adolescents. In this section, the findings are discussed in light of the existing literature. In the present study, higher levels of social media addiction were found to be significantly associated with higher levels of ghosting. According to what Kudubes and Efe’s [21] found, social media addiction negatively affects adolescents’ daily activities and can disrupt their routine lives and social interactions. This situation is supported by similar trends in our study. Kucuk and colleagues [22] assert in their study that that rise in social media use among adolescents correlates with a heightened risk of social media addiction, cyberbullying and victimisation also increases. This study shows similarities with the aforementioned research. Morey and colleagues [23] found that young people use digital communication technologies intensively and that ghosting behaviour is more common in these environments. These findings align with the results acquired in our investigation. In our study, a significant relationship was found between emotional intelligence and ghosting levels. Lower emotional intelligence scores were associated with higher ghosting scores. Piqueras and colleagues [24] emphasised in their study that emotional intelligence may protect one from against psychopathology in adolescents. This research illustrates the beneficial impact on mental health that comes from helping adolescents develop emotional intelligence. This aligns with the conclusions of the research by Piqueras et al. [24] Poulou [25] discovered that students with elevated emotional intelligence and robust social skills were less prone to display emotional and behavioral challenges. The study in question revealed similar outcomes. Schutte and Malouff [26] reported in their study that elevated emotional intelligence correlates with enhanced positive affect and life satisfaction, meanwhile diminishing negative affect. In the context of the findings of this study, emotional intelligence may contribute to individuals’ ability to manage their emotional processes and interpersonal relationships. This may help explain why lower levels of emotional intelligence were associated with higher levels of ghosting in our study. Our study also found that higher social media addiction scores were significantly associated with higher ghosting scores. This finding is consistent with studies highlighting the influence of digital communication environments on interpersonal relationship processes Halversen and colleagues [27]’ interviews with 419 women who had recently rejected someone they met on social media or dating platforms reveal that the way a relationship begins is tight ties with the concept of sudden disappearance, and that relationships that begin on social media are more likely to end with a disappearance strategy. This study’s findings are in agreement with our own. Forrai and colleagues [28] also found that individuals who met on a dating app most frequently chose the strategy of suddenly disappearing as their relationship termination strategy. Similarly, our study found that social media addiction and emotional intelligence levels were significantly associated with ghosting behavior. However, the fact that the hierarchical regression model explained 12.2% of the variance suggests that, in addition to social media addiction and emotional intelligence, other factors may also contribute to ghosting behavior. Consistent with this finding, the existing literature indicates that ghosting behavior is associated with attachment styles, communication processes, conflict avoidance tendencies, and various psychological characteristics [6,10]. Nevertheless, ghosting is considered a multidimensional behavior influenced by the characteristics of digital communication environments, the context of the relationship, and the dynamics of interpersonal interactions [5,10]. The machine learning analyses supported the findings of the hierarchical regression models. While the regression analysis demonstrated that social media addiction and emotional intelligence were significantly associated with ghosting, the machine learning analyses provided further insight into the relative importance of these variables in predicting ghosting behavior. The SHAP analysis revealed that social media addiction made the greatest contribution to the prediction of ghosting behavior, whereas emotional intelligence contributed at a secondary level. This finding suggests that social media addiction is the most influential predictor of ghosting behavior among the variables examined. Conclusions In this study, social media addiction and emotional intelligence were found to be significantly associated with ghosting behavior among adolescents. The findings indicate that higher levels of social media addiction and lower levels of emotional intelligence are associated with higher levels of ghosting. Based on these results, it is recommended that educational and intervention programs aimed at promoting healthy social media use habits and enhancing emotional intelligence skills during adolescence and young adulthood be expanded. Furthermore, longitudinal studies involving adolescents are recommended to better understand changes in ghosting behavior over time and to identify its potential determinants. Strenghts and limitations The present study has several important strengths. First, to the best of our knowledge, it is among the limited number of studies that directly examine social media addiction, emotional intelligence, and ghosting behavior together among adolescents. Second, the use of a sufficiently large sample of 425 adolescents enhances the reliability of the findings. In addition, the application of machine learning approaches enabled the relationships among the variables to be examined from different analytical perspectives. In particular, the SHAP analysis allowed for the determination of the relative importance of variables in predicting ghosting behavior. However, the study also has several limitations. First, due to its cross-sectional design, the findings do not permit causal inferences. Second, because the data were collected using selfreport measures, social desirability bias and recall bias may have affected the results. Third, participants were recruited through online distribution and a snowball sampling method. Consequently, the sample may not be representative of all adolescents in Türkiye, and the possibility of selection bias should be considered. Furthermore, the hierarchical regression model explained only 12.2% of the variance in ghosting behavior. This suggests that various individual, interpersonal, and psychosocial factors not assessed in the present study may also contribute to ghosting behavior. Finally, although the machine learning analyses supported the regression findings, the resulting models require validation in different samples.