Feature Selection Using Chi-Squared Feature-Class Association Model for Fake Profile Detection in Online Social Networks

Protecting online social networks requires effective detection of fake profiles, combined with feature selection to identify and mitigate fraudulent accounts, thereby improving classification accuracy in the detection task. Several existing methods have been proposed for identifying fake profiles, incorporating different features. However, the true efficiency of the classification model relies on selecting suitable features for classification. A standard statistical analysis, the chi-square test compares the distribution of two populations. Due to its advantage, the proposed study suggests a chi-squared feature-class association model for feature selection in the context of fake profile identification in social networks. This approach evaluates the relevance of associations between features and target class, enabling the identification of vital features for classification. Each feature in the dataset is analyzed for its chi-square statistic in relation to each target class, and the features with the highest chi-square values for each class are prioritized. The proposed method selects the top-scored features from each class based on their chi-square values. Further, the ability of the model in detecting fake profiles is improved by the use of the class proportion in determining the number of selected features within each class. The research uses three public datasets for an in-depth analysis. The proposed feature selection method achieves 91.9%, 89.3%, and 98.5% in average accuracy for the datasets of Facebook, Instagram, and Twitter respectively. Besides, the study reveals that the proposed model has a considerably highly effective and has a high-performance margin as compared to any other of its related competitive methods. This enhanced performance of the proposed model demonstrates a rise in the accuracy of detecting fake profiles.