Predictive Analysis for the Early Detection of Depression in Adolescents Based on Social Media Usage Patterns

Figure Ilustration AI

FORMOSA NEWS - Medan - Excessive social media use among teenagers is a growing public health concern, frequently linked to rising rates of anxiety, stress, and emotional instability. To address this challenge, a groundbreaking 2026 study by researchers at Universitas Sumatera Utara (USU) has successfully developed a predictive machine learning model to detect early-stage depression in adolescents by analyzing their digital behavior and daily lifestyle patterns. The research, conducted by Genesis Sembiring Depari and Julpan Daniel Simatupang from Universitas Sumatera Utara, leverages advanced data analytics to identify high-risk individuals before severe clinical symptoms manifest. Published in the Formosa Journal of Computer and Information Science, this study reveals that sleep duration and daily social media screen time are the most critical predictors of mental health vulnerabilities in youth. The findings offer a scalable, data-driven alternative to traditional psychological assessments, which are often costly and resource-intensive.

The Digital Dilemma: Screen Time vs. Mental Well-Being
As digital platforms become deeply embedded in adolescent life, uncontrolled exposure to virtual environments presents significant psychological risks. Teenagers are highly vulnerable to digital distress because they are undergoing critical phases of emotional, cognitive, and social development. Constant online engagement often fosters unhealthy social comparisons, exposes youth to cyberbullying, and reduces physical activity all of which act as catalysts for depressive disorders. Furthermore, social media addiction has emerged as a major modern challenge. Adolescents experiencing high levels of digital dependency frequently suffer from poor academic performance, impaired focus, and severe sleep deprivation. To explore these behavioral relationships, Genesis Sembiring Depari and Julpan Daniel Simatupang analyzed a secondary dataset containing 1,200 unique adolescent records. This comprehensive dataset evaluated multiple daily behavioral and psychological variables, including:

• Daily social media usage duration.
• Social media platform preferences.
• Nightly sleep duration.
• Self-reported stress and anxiety levels.
• Screen addiction tendencies.
• Academic performance and physical activity levels.

The analysis confirmed a strong positive correlation between prolonged daily screen time and symptoms of depression. Crucially, excessive social media use, especially late at night, directly causes shorter sleep duration. This lack of adequate rest severely impairs emotional regulation, making adolescents significantly more susceptible to mental health struggles.

Machine Learning Performance: Decision Tree Algorithm Outperforms the Rest
To establish an automated early detection system, the Universitas Sumatera Utara researchers evaluated four prominent machine learning algorithms: Logistic Regression, Decision Tree, Random Forest, and Support Vector Machine (SVM). The dataset was divided using a standard 80:20 ratio for model training and testing. The performance of each predictive model was evaluated across several key classification metrics:

Machine Learning Model	Accuracy	Precision	Recall (Sensitivity)	F1-Score	Area Under Curve (ROC-AUC)
Logistic Regression	98.75%	100.00%	50.00%	66.67%	98.79%
Decision Tree	99.17%	83.33%	83.33%	83.33%	91.45%
Random Forest	97.92%	100.00%	16.67%	28.57%	99.64%
Support Vector Machine	97.50%	0.00%	0.00%	0.00%	98.86%

While models like Logistic Regression and Random Forest yielded exceptional overall accuracy rates, they demonstrated low sensitivity (recall) in identifying actual positive cases of depression. This means they excel at identifying healthy individuals but frequently miss those who are genuinely at risk. In contrast, the Decision Tree algorithm achieved the most balanced and reliable performance. With an impressive recall and F1-score of 83.33%, the Decision Tree model proved to be the most effective tool for detecting active depressive symptoms in adolescent behavior datasets.

Real-World Impact: Designing Better Interventions
The integration of predictive analytics into public health and education provides a powerful tool for early intervention. Historically, assessing adolescent mental health has relied on manual, subjective, and slow psychological evaluations. By deploying automated machine learning models, schools, healthcare providers, and parents can monitor behavioral risk factors seamlessly and intervene before a crisis occurs. Through their research at Universitas Sumatera Utara, Genesis Sembiring Depari and Julpan Daniel Simatupang emphasize that addressing adolescent depression requires looking beyond clinical environments. Digital platform developers, educational institutions, and policymakers must collaborate to design environments that support digital well-being. Practical prevention programs should prioritize healthy screen-time limits, promote sleep hygiene, and teach digital literacy to help teenagers navigate online spaces safely.

Author Profiles
Genesis Sembiring Depari, S. Pd, MBA, Ph.D is a computer science researcher affiliated with Universitas Sumatera Utara. His research interests focus on predictive business analytics, machine learning applications, and data-driven mental health monitoring systems.
Julpan Daniel Simatupang, S. Pd. . is an academic and researcher at Universitas Sumatera Utara. He specializes in information technology systems, machine learning algorithms, and the integration of predictive modeling in health and education sectors.

Source
Genesis Sembiring Depari, Julpan Daniel Simatupang. Predictive Analysis for the Early Detection of Depression in Adolescents Based on Social Media Usage Patterns. Formosa Journal of Computer and Information Science (FJCIS). Vol. 5 No. 1 2026. Hal 215-228.

DOI : https://doi.org/10.55927/fjcis.v5i1.16904
URL: https://journal.formosapublisher.org/index.php/fjcis