AMLFN-AD:Adaptive multi-level integrated fusion attack detection framework for intelligent building systems

Abstract

With the development of emerging information technologies such as Internet of Things (IoT), cloud/edge computing and big data, the traditional building has been promoted in the direction of intelligence and become an indispensable part of the smart city nowadays. A higher level of intelligence and automation in the building can further meet people’s diverse needs. However, the lack of network security protection can pose a serious threat to intelligent building systems and their users. Due to the complex and changeable network environment, network attack detection may face the difficulty of data and model overload and unsatisfactory detection accuracy. Although many network security technologies have been proposed, few of them aim at intelligent building systems. In view of this, we propose an adaptive multi-level integrated fusion network attack detection framework (AMLFN-AD) to detect network attacks for intelligent building systems. In the first level, an efficient classification model is used to quickly distinguish attack from normal samples. Then, attack and misclassified normal samples are reclassified in fine granularity with an adaptive ensemble model in the second level. We adopt a hybrid model selection method to adaptively choose base classifiers from a pre-trained model pool for the ensemble model. Moreover, oversampling and undersampling techniques are combined to reduce the impact of data imbalance problem. A series of experiments against other comparative methods are conducted on three datasets, and the obtained results show that AMLFN-AD can achieve superior performance.