Anomaly detection is a process in which abnormal behavior or detection of abnormal samples is identified by comparing them with normal data samples. In these years, there has been an interesting focus on anomaly detection for more caution, in various fields such as medical care, banks and payment outliers, financial monitoring and detection of network attacks. In this research, a new anomaly detection approach called AdaDL_HPO-SVDD_HPO is introduced to solve the challenge of uncertain data. This method will use both normal and abnormal examples to create sparse representations through dictionary learning during training. In this research, Support Vector Data Description (SVDD) is integrated in the framework to create a minimum hypersphere for anomaly detection in experimental data. On the other hand, AdaBoost, as an ensemble learning method, has been used to create a robust classifier by combining weaker classifiers. AdaBoost algorithm is very sensitive to noise in data and hyperparameter changes. Therefore, to solve this problem and to increase accuracy in anomaly detection, the hyperparameters of both SVDD and AdaBoost algorithms have been optimized. The simulation results show that the proposed method succeeds in anomaly detection with 95.2% accuracy.
