A fascinating area of study, smart facial expression recognition has been proposed and used in a number of domains, such as human-machine interactions, defense, and health. In order to improve algorithm prediction, researchers in this field are concentrating on methods for encrypting, decoding, and erasing face expressions. An Internet user’s most prominent emotional state whether that state might be happiness or sadness or anger or anxiety or boredom or neutrality could be plugged into a machine-learning-based tool in this work. We used a dataset of 1,000 samples and ten features, including daily usage time, posts, likes, and comments, to train, and evaluate five models (Random Forest, KNN, SVR, XGB, and Neural Network). KNN outperformed the best with a test accuracy of 95.15% and validation accuracy of 78%. Furthermore, a simple GUI based on Python and Java Swing was designed for real-time emotion prediction. It provides a practical utilize of machine learning classifying social media usage and conjugate these subtle emotional affect. The results basically proves how KNN works with hierarchical datasets and can work toward creating consciousness about wellbeing on social media. In future, it can be recommended to add various other features along with data collection in real-time.