An Intrusion Detection System Using a Machine Learning Approach in IOT-based Smart Cities

Abstract

For a long time, the digitization of all aspects of life in current cultures is seen as a procured gain. In any way, the computerized world is noticeably flawed and numerous risks and dangers are present as in the terrestrial land. People's daily life has changed due to the quick and advanced level of improvement in smart cities. The most important problem that needs to be looked upon is citizens' life, security and privacy issues. The use of Deep Learning (DL), a subcategory of Machine learning (ML) has excelled in the field of smart cities. So, the following stages in this paper bring an effective intrusion detection system using deep learning. a) Data collection from standard datasets such as GPRS, CIDDS001, as well as UNSW-NB15 contains various types of attacks, these will be given for b) Preprocessing, for eliminating anomalies using missing value removal, and normalization techniques. Then from those data, quintessential features are extracted using Autoencoder (AE) and then from those several features, d) feature selection for selecting and mostly removing timestamps from attack dataset using Random Forest (RF) and finally for e) prediction with help of Restricted Boltzmann Network (RBN). Experiment evaluation states that proposed model (RF-RBN) performed better over various state-of-art models under various measures (accuracy:0.95, sensitivity:0.96, specificity:0.97, detection rate:0.95).