Anomaly Detection in Edge Computing using Deep Fuzzy Hypersphere Neural Network Learning Model on NSL-KDD Dataset

Abstract

IoT devices have been extensively utilized on numerous smart applications such as smart city, healthcare, and Industry. Since IoT devices possess tiny computing power and not capable to compute large volumes of data, in spite of the advantages of IoT, it also possesses inherent drawbacks like latency, bandwidth limitation, reliability concerns, and security risks. Edge computing counteracts these drawbacks by processing the data locally and implemented for processing this much huge sensors data on cloud. The Edge will process the data closer to where it is created so that processing may be accelerated and latency can be reduced, again in Edge Computing a variety of irregularities in data generation are generated by the increasing heterogeneity and complexity of edge devices due to their limitations. Anomaly detection is a crucial task in edge computing systems, where identifying unusual or deviant patterns of data is essential to ensuring system security and reliability. An original Deep Fuzzy Hypersphere neural network learning model (DFHNNLM) is proposed in this paper for effective anomaly detection in edge computing tasks. The proposed method outperforms current state-of-the-art for anomaly detection with existing deep learning techniques. Proposed model is suitable for any anomaly dataset like ECG5000, NSL-KDD. According to the experimental results, the DFHNNLM outperforms both deep learning and conventional machine learning methods in anomaly detection, achieving improvements in F1-score, accuracy, precision, and recall.