Abstract:
The fast emergence of short-form video platforms such as Instagram Reels, TikTok, and Facebook Reels has changed the face of digital participation especially among the younger generations and is already shaping expectations around such content. This study will investigate short-form social media reels addiction, and risk of mental illness, specifically, anxiety, depression, stress, and sleep problems in adolescents and young people ages 13–25 years old, in Bangladesh.A cross-sectional survey of 135 active reels users was conducted, collecting demographic information, usage, and mental health self-reported indicators. Five supervised machine learning algorithms were developed (Logistic Regression, Random Forest, Naive Bayes, k-Nearest Neighbors and Multilayer Perceptron) to predict mental health risk from the data overall, featuring summary behavioral and demographic information. The model evaluation metrics used were accuracy, AUC, precision, recall, F1 score, and Matthews Correlation Coefficient. The Logistic Regression model was the most powerful model, achieve the highest accuracy (94.1%) and AUC (0.986), while Random Forest achieved the second highest accuracy (92.6%) and AUC (0.980), but both had low false-negative rates, which is ideal for recognizing these risks early.The primary predictors include spending over two hours per day viewing social media reels, viewing reels late in the evening, and developing social comparison behaviours. Social media users are already connected to mental health problems, but they found that even a small amount of usage and risky behaviours not related to being an addict could worsen mental health vulnerability for some patients.
