Abstract
This paper presents an approach to detecting fraudulent user profiles on online social networks. The idea is to build a model(s) based on all multiple types of data gathered from online social networks. Here we build models on the Twitter dataset, where we compose data, such as user account information, user connections, and the content that the users are producing. This enables us to create models that are robust, almost data type and content independent. Combining multiple models for the same purpose facilitated the creation of a final model that is self-reliant and adaptable to different data. The results are approximately perfect and are very efficient in terms of detecting fraudulent user profiles. Consequently, this model has the capacity to detect fake or spam user profiles, which can lead to cleaner Internet space.