Non-Trusted user Classification-Comparative Analysis of Machine and Deep Learning Approaches

A cloud environment in any context is subject to threats due to various non-trusted users. Non-trusted users might be the new users or existing trusted users who turn into non-trusted users. Classification of these users is needed in all scenarios. It has also been observed that the history of user behavior plays a key role in identifying non-trusted users. Such data would be high in volume and machine learning models are found to better suit them in classifying the non-trusted users. Instead of using the real-time data more similar data from the insider threat dataset is employed in the work. Since the features in the insider threat dataset cannot be directly used in the machine learning models, new features were created from them and a new dataset is created. Four different machine learning classifiers, Logistic Regression, Decision Tree, Random Forest Classifier, and K-Nearest Neighbor are used for classification. In addition to this, a deep neural network model is also created. The results are compared and it is found that the Deep Neural network model performs well than the other models with an accuracy of 88% and an AUC value of 0.856. the sigmoid activation function is used in the output layer of the deep artificial neural networks and hence it produces probability values that reflect the trust value of the users.