Abstract
Abstract: Detection of anomalies and non-anomalies in social networks is critical. The unusual and mischievous events occurring in these networks are due to anomalies. Graph metrics inspired researchers to track anomalous behaviors using social networks' graphical structure. Determining graph metrics to define anomalies and non-anomalies has been a significant hurdle for researchers. This paper details the development of a machine learning system to detect anomalies and non-anomalies based on graph metrics (Betweenness Centrality BC, Closeness Centrality (σcci), Eigenvector Centrality (σecj), Clustering Coefficient (σclcj)) with predictive models to classify the anomalies and non-anomalies, such as Support-Vector-Machine (SVM), Logistic-Regression (LR), and Neural-Network Multi-layer Perceptron (NN-MLP) on Facebook. Experiments were carried out as well for evaluation purposes. The results were assessed based on performance appraisal metrics such as Precision, Sensitivity, and the F1-score. The experimental results confirmed that the Support-Vector-Machine is superior to the other models.