Social Media User’s Safety Level Detection through Classification via Clustering Approach

Abstract

Social media has a significant impact on our daily life, and the popularity is increasing rapidly because of the ability to be attached to people around the world and share feelings, photos, videos, etc. So, it bears a high-security concern. However, most of the social media user does not know the security level of their account, including what features of social media should consider if the account is in a risk situation. The posting, friendship, etc. sometimes brings unfortunate events like identity theft, sexual harassment, cyber-crime, etc. To overcome such kind of unexpected issues, this research proposes a classification via clustering algorithm based predictive model by which one can know his safety level in the social media. A dataset is formed through a closed-ended questionnaire. Essential features are selected via gain ration method as high dimensional data is costly to train a model. An unsupervised algorithm, hierarchical clustering, cluster the users into three groups that are labeled for further analysis. The various classification algorithm is chosen to train the predictive model. From the model evaluation result, “Logistic Regression” predicts the safety level of a social media user with high accuracy. So, this model will bring an extra dimension in social media user account safety.