Intelligent deep learning framework for cache pollution threatening detection using named data networking

Malicious management of the caching process disrupts data retrieval, causing an unavoidable threat known as a
cache pollution attack (CPA), which reduces network performance, increases information retrieval time, and
decreases the cache hit rate. Due to the importance of CPA detection in NDN, this study introduces the enhanced
deep learning (EDL) model to improve data integrity and security. The EDL model integrates the idealogy of
recurrent neural networks and a memetic optimization algorithm to eliminate malicious user participation in
networks. During the analysis, the CICIDS 2017 dataset was utilized due to its high-dimensional data and
coverage of diverse attacks. The gathered details are processed by traffic preprocessing blocks that remove

irrelevant, missing, and redundant information by considering specific filtering conditions. Then, the normal-
ization process is applied to mitigate the overfitting issue and reduce computational complexity. The normalized

inputs are fed into the recurrent layer, which identifies the relationship between features such as temporal patterns,
cache hit rates, and frequency of access requests. According to the relationship, the malicious and legitimate users
are identified with maximum accuracy. In addition, memetic operators such as selection, crossover, and mutation
parameters are utilized to balance the stability between the features. The effective selection parameter ensures
optimal results while managing the cache in NDN, achieving a 3.2% error rate and a cache hit rate of 98% to 99%.