Improved detection of Facebook anomalies and abnormalities using graph-based and machine learning techniques





Journal title

AIP Conference Proceedings


Ramzi Hamid Elghanuni


ABSTRACT The explosive growth of online-Social-Networks and the vast number of their subscribers’ data has played a major role at this time to the trend toward interest to detect suspicious and anomalous behaviours on this OSNs. Moreover, the attracted attackers and impostors were intercepted personal data, share false news and promote malignant activities. Thus, the researchers were encouraged by the graphical structure of social networks to use different graph metrics and classification algorithms to detect anomalous activities. However, some of those classification algorithms have been a negative effect or no effect on the end-results, nor do they always achieve satisfactory results by using standalone classification algorithms. This article introduces a new algorithm, called PCA-NN which suggest efficient detection of anomalies and non-anomalies for Facebook. Also, the feature selection and Principal Component Analysis (PCA) techniques have applied. furthermore, classification algorithms for machine learning have used to assess the anomalies and non-anomalies such as supporting vector machine (SVM), neural network (NN), logistic regression (LR) and our built algorithm PCA-NN. Lastly, a minimum number of features have used by the proposed algorithm (PCA-NN) to identify the anomalies on online social networks with the highest accuracy and sensitivity. REFERENCES

Publisher's website