Social phobia is one of the anxiety which affects students worldwide, by affecting their performance in academics as well as their interaction with society. To identify social phobia accurately, we create an advanced machine learning-based network model “TwinNet Architecture” by utilizing the physiological and eye-tracking data. To identify the model working, we utilized three different publicly available datasets. Through the learning of different features, the network identifies the difference between the two labels, and an accuracy of 81% is achieved through the novel architecture, and the possible future scope of the model is discussed in this article.