Deep Learning to Improve False Data Injection Attack Detection in Cooperative Intelligent Transportation Systems

This work proposes a local-global fuzzy-clustering feature extraction scheme for detecting False Data Injection Attacks (FDIA). In this scheme, the data undergo several pre-processing steps including missing values imputation based on the local and global fuzzy-clustering correlation approach. There are four main components of the proposed method: i) data acquisition, ii) standardization, iii) normalization, and iv) imputation. To evaluate the performance of this scheme, the NGSIM dataset (described below) was used. This dataset contains data acquired from the environment using a set of sensors that collect data from the neighbouring vehicles. The results show that the accuracy of models trained using said features extracted by the proposed scheme was higher than those proposed by the related studies. This indicates that the local-global fuzzy clustering data imputation approach proposed by this study can estimate the missing values better than existing techniques based on an exhaustive literature review.