Practical autoencoder based anomaly detection by using vector reconstruction error

Abstract Nowadays, cloud computing provides easy access to a set of variable and configurable resources based on user demand through the network. Cloud services are available common internet protocols network standards. In addition unique benefits computing, insecure communication attacks networks cannot be ignored. There several techniques for dealing with attacks. To this end, anomaly detection systems widely used as an effective countermeasure against anomalies. The anomaly-based approach generally learns normal traffic patterns in various ways identifies Network have gained much attention intelligently monitoring using machine learning methods. This paper presents efficient model autoencoders networks. autoencoder basic representation data its reconstruction minimum error. Therefore, error is or classification metric. addition, detecting from data, types has also been investigated. We proposed new by examining autoencoder’s method Unlike existing autoencoder-based that consider all input features single value, we assume vector. enables our use every feature further propose multi-class structure classify CIDDS-001 dataset commonly accepted literature. Our evaluations show performance improved considerably compared ones terms accuracy, recall, false-positive rate, F1-score metrics.