With the growing number of users on online social networks (OSNs), concerns about excessive usage and social media addiction are becoming increasingly serious. Addiction to social networks can negatively impact both physical and mental health over time. Instagram, as one of the most popular platforms, occupies a significant amount of users’ time. As a result, developing methods for the automatic detection of social network addiction is essential. However, such methods also raise important privacy concerns. This paper demonstrates that it is possible to predict a user’s likelihood of social media addiction solely based on their online behavior. A questionnaire based on the Bergen Social Media Addiction Scale (BSMAS) was designed and distributed to Iranian Instagram users. Profile and activity data were then collected from the respondents’ pages and analyzed using machine learning algorithms, including Genetic Algorithm, Multi-Layer Perceptron (MLP),which is a type of neural network, and Support Vector Machine (SVM). Additionally, an autoencoder was used to extract key behavioral features, and its performance was compared with models trained on the original features. The results showed that the SVM with original features and the MLP with autoencoder-extracted features achieved the highest accuracy in classifying users as addicted or non-addicted. The findings indicate that social network addiction can be inferred from behavioral data alone, highlighting potential privacy risks. While behavioral analysis techniques are effective for addiction detection, they could also be misused to monitor users without their knowledge or consent.