Abstract
Modern-day social media is one of the most used platforms by millennials for sharing personal, and professional events, thoughts & other entities. These entities include photos, texts, videos, locations, meta data about other users, etc. Thus, securing this content from fake-users is of utmost importance, due to which a wide variety of techniques are proposed by researchers that includes but is not limited to, deep learning models, high density feature processing models, bioinspired models, etc. But these models are either highly complex, or require large user-specific datasets in order to improve their detection capabilities. Moreover, most of these models are inflexible, and cannot be scaled for large social networks with multiple parameter sets. To overcome these issues, this text proposes the design of a novel fusion model to identify fake profiles from multimodal social media datasets. The proposed model initially collects multimodal information about users that includes the presence of profile pic, username length ratios, number of words in the full name, length of their personal description, use of external URLs, account type, number of posts, number of followers & following users, etc. These information sets are pre-processed via a Genetic Algorithm (GA) based feature selection model, which assists in the identification of highly variant feature sets. The selected feature sets are classified via a fusion of Naïve Bayes (NB) Multilayer Perceptron (MLP), Logistic Regression (LR), Support Vector Machine (SVM), and Deep Forest (DF) classifiers. Due to a combination of these classifiers, the proposed model is capable of showcasing an accuracy of 98.5%, precision of 94.3%, recall of 94.9%, and F-measure score of 94.7% across multiple datasets. Due to such a high performance, the proposed model is capable of deployment for a wide variety of social media platforms to detect fake profiles.