Research

  • Predicting Non-suicidal Self-injurious Behavior and Suicidal Ideation Based on Mental Illness among University Students Using Machine Learning Techniques: A Study from Bangladesh

[Under peer review in the Scientific Reports (First Author)]

Abstract

Background: Non-suicidal self-injury (NSSI) and suicidal ideation (SI) represent significant mental health challenges among university students. In low- and middle-income contexts like Bangladesh, there is limited understanding of how these behaviors differentially affect students with and without mental illness. This study addresses these gaps by investigating the prevalence and risk factors of NSSI and SI, with stratified analyses by mental illness status, and by employing advanced machine learning (ML) techniques to predict these behaviors. Methods: This cross-sectional study recruited 1,401 university students between December 2024 and January 2025. Data was collected via a self-administered questionnaire assessing socio-demographics, psychological factors, and body image. Traditional statistical analyses, including chi-square tests and logistic regression, were conducted in SPSS 27. Six supervised machine learning algorithms were applied in Python using an 80/20 train-test split, SMOTE for class imbalance, and 5-fold cross-validation. Model performance was evaluated by accuracy, precision, F1 score, log loss, and ROC analysis. Results: The prevalence of NSSI and SI was 21.4% and 17.2%, respectively. Behavioral and psychosocial factors, including smoking, depressive and anxiety symptoms, cyberbullying, physical inactivity, and eating disorders, were consistently associated with both NSSI and SI. In ML models, CatBoost achieved the highest performance for NSSI prediction (accuracy: 79.9%, F1-score: 74.9%), while Random Forest performed best for SI (accuracy: 83.5%, F1-score: 79.3%). SHAP analysis identified current smoking, eating disorders, depression, and cyberbullying as leading predictors. Conclusions: This is the first study in Bangladesh to integrate both traditional and ML-based approaches for identifying risk factors and classifying NSSI and SI among university students, with attention to mental illness subgroups. The findings highlight the high prevalence and complex etiology of self-harming behaviors and demonstrate the value of ML for early detection and risk stratification. ML-based screening tools hold significant potential for enhancing targeted interventions and informing suicide prevention strategies in resource-limited academic settings.

Keywords: Non-suicidal self-injury, Suicidal ideation, Self-harming behavior, University students, Psychological risk factors, Bullying and mental health.

  • Predicting Mental Health Outcomes in University Students Using Machine Learning Techniques: A Study from Bangladesh. [Under peer review in the Journal of Biosocial Science (First Author)]

Abstract : Mental health challenges are a growing global public health concern, with university students particularly vulnerable due to academic, social, and behavioral stressors. In Bangladesh, limited research has explored mental health outcomes using both traditional statistical and advanced machine learning (ML) approaches. This study assessed the prevalence of depression, anxiety, and stress among university students and identified associated risk factors using an integrated analytical framework. A cross-sectional survey was conducted among 1,485 students from two Bangladeshi universities, and collected data on sociodemographic, behavioral, and health-related factors. Depression, anxiety, and stress were measured using the validated Bangla version of the DASS-21. Statistical analyses included chi-square tests and binary logistic regression, while six ML models were applied to predict mental health outcomes and assess feature importance. The prevalence of depression, anxiety, and stress was 56.9%, 61.0%, and 32.1%, respectively. Female gender, cigarette smoking, unfriendly family relationships, academic discipline, and year of study were identified as significant risk factors. GBM showed the highest predictive accuracy for depression (54.64%) and anxiety (59.35%), while SVM performed best for stress (67.45%). CatBoost achieved the lowest log loss across outcomes, indicating strong model stability. Feature importance analyses identified cigarettes smoking and family relationship quality as key predictors. Findings suggest mental health problems are widespread among university students. Integrating ML with traditional methods enhances understanding of complex risk patterns and supports the development of culturally appropriate interventions in low- and middle-income settings.

Keywords: Mental Health; Machine Learning; University Students; Gradient Boosting Machines; SHAP.

  • Problematic social media use among university entrance test-takers: Prevalence, psychosocial factors, and a mediation-moderation model [Under peer review in the Addictive Behaviors (First Author)]

 

Abstract: Social media has become integral to daily life, but problematic social media use (PSMU) is an emerging public health concern. Few studies have specifically examined PSMU among university admission test-takers. This study aimed to investigate the prevalence and predictors of PSMU among university entrance exam candidates in Bangladesh. A cross-sectional study was conducted in February 2025, involving 1163 students preparing for university admission tests. Data on sociodemographic, admission-related factors, mental health symptoms, perceived stress, social support, and PSMU were collected. Data analysis involved Chi-square tests, logistic regression, and structural equation modeling (SEM) using IBM SPSS 26 and R (lavaan package). The prevalence of PSMU was 21.2%. Logistic regression analysis revealed that social media use duration, cigarette smoking, fracture in body parts, depression (OR = 1.60, 95% CI = 1.10–2.34) and high stress (OR = 1.65, 95% CI = 1.03–2.64) had significantly increased odds of developing PSMU. Participants with low and moderate social support had higher risks of PSMU (OR = 1.72 and 1.51, respectively). SEM analysis indicated that anxiety (β = 0.37, p = 0.009) and stress (β = 0.27, p < 0.001) had significant direct effects on PSMU, whereas depression did not directly influence PSMU. Social media use duration significantly mediated the effect of stress on PSMU (indirect β = 0.089, p = 0.003). Anxiety, stress, and social media usage duration contributes to PSMU. These results inform targeted interventions to mitigate PSMU behaviors and support mental health in this vulnerable group.

Keywords: Problematic social media use; psychological distress; depression; anxiety; stress; perceived social support

  • Associations of psychosocial and body features related factors with body image disturbance and body dysmorphic disorder among university students [Under peer review in the Scientific Reports]

Abstract

Background: Body image disturbance (BID) and body dysmorphic disorder (BDD) are significant mental health concerns, particularly among young adults. Both conditions are linked to various psychological, social, and behavioral factors; however, research in this area remains scarce in Bangladesh. This study aims to assess the factors of BID and BDD among university students in the country. Methods: This cross-sectional study utilized a convenience sampling technique and surveyed 1,401 university students between December 2024 and January 2025. Data were collected using a structured and self-administered questionnaire assessing socio-demographics, psychological factors, and body image concerns. Statistical analyses, including t-tests, ANOVA, and multiple linear regression, were employed to identify associated factors, with data analysis performed using IBM SPSS 27. Results: The mean BID score was 12.0 (±5.39; range: 1–35), while the mean BDD score was 3.37 (±2.76; range: 0–12). Concerns related to hair (32.0%), acne (29.1%), and scar marks (17.8%) were the most commonly reported sources of discomfort. Gender-based analyses revealed that female students had significantly higher BID and BDD scores than males (12.70 ± 5.6 vs. 11.33 ± 5.09, p<0.001; and 3.53 ± 2.92 vs. 3.22 ± 2.60, p=0.036, respectively). Additionally, students reporting discomfort with body features (e.g., acne, skin tone, teeth), self-harm, suicidal ideation, bullying, or cyberbullying experiences, as well as those with eating disorders, depression, or anxiety, demonstrated significantly higher BID and BDD scores. The regression model accounted for 24.9% and 21.7% of the variance in BID and BDD scores, respectively. Conclusion: The findings highlight the burden of BID and BDD among Bangladeshi university students, particularly among females and individuals experiencing psychological problem or bullying. Targeted interventions, including mental health support programs, awareness campaigns against unrealistic beauty standards, and anti-bullying policies, are crucial to addressing these concerns.

Keywords: Body Image; Body Image Dissatisfaction; Body Part Discomfort; Obsessive Compulsive Disorder; Bangladesh.

  • Associations of psychosocial and sociodemographic factors with traditional bullying and cyberbullying among university students: A cross-sectional study

Abstract

Background: With the escalation of digitalization, both cyberbullying and traditional bullying have emerged as significant risk factors for the physical and mental health of adolescents in Bangladesh, consistent with global patterns. Despite global attention, limited research addresses the combined burden of these phenomena in Bangladeshi universities. This study aims to explore the prevalence, psychological impacts, and key demographic factors associated with both types of bullying among university students in Bangladesh. Methods: This cross-sectional study utilized a convenience sampling technique and surveyed 1,397 university students between December 2024 and January 2025. Data were collected using a structured and self-administered questionnaire assessing socio-demographics, psychological factors, and body image concerns. Statistical analyses, including Chi-square tests and binary logistic regression, were employed to identify associated factors, with data analysis performed using IBM SPSS 27. Results: Our findings revealed that 41.5% of university students reported experiencing traditional bullying, whereas 24.3% encountered cyberbullying, with a slightly higher prevalence among females. Female students demonstrated significantly higher odds of experiencing both traditional (AOR=1.4, p=0.009) and cyberbullying (AOR=1.4, p=0.026). Overweight status was associated with increased risks of traditional (AOR=1.61, p=0.049) and cyberbullying (AOR=2.15, p=0.009), while obesity further elevated the risk of cyberbullying (AOR=2.74, p=0.018). Smoking behavior emerged as a strong predictor for traditional (AOR=2.05, p<0.001) and cyberbullying (AOR=2.48, p=0.001). Depression (AOR=1.35, p=0.017) and suicidal ideation (AOR=2.24, p=0.001) were associated with traditional and cyberbullying, respectively. Eating disorders predicted cyberbullying (AOR=1.77, p<0.05), while parental education and income were exclusively linked to traditional bullying. Conclusion: The study highlights significant rates of bullying, recommending integrated university strategies encompassing policy reforms, mental health support, and peer-driven programs to mitigate bullying’s detrimental psychosocial and behavioral consequences.

Keywords: Traditional Bullying, cyberbullying, university Students, psychosocial Factors, Bangladesh

 

 

  • On-going projects:
  • Exploring the associations between adverse childhood experiences (ACEs) with sleep duration and quality: A cross-sectional study

Current Status: Following the completion of data analysis, the study has entered the manuscript development phase.

 

  • The Impact of Academic Performance, Perceived Stress, Gastroesophageal Reflux Disease, and Obesity on Night Eating Syndrome and Sleep Health Among University Students

Current Status: Research methodology has been finalized, and data collection is currently in progress.

[Data collection (via google form) link: https://forms.gle/6nQPApHb8Wwu9rb49 ]