A Comprehensive Analysis of Machine Learning Models for Real Time Anomaly Detection in Internet of Things

Sanjith, S.; Raj, E. George Dharma Prakash

doi:10.26438/ijcse/v6i11.932937

Cited by 1 publication

(1 citation statement)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The aggregate measures; accuracy and F1-Score was also observed to exhibit high values (>0.98), exhibiting the enhanced prediction levels of the proposed model. A comparison of the proposed DTRAD model was performed with recent anomaly detection models like RampLoss [11] and the DBSEM model [25] is shown in figure. Comparisons were performed based on accuracy, FPR, Precision and F1 Score. It could be observed that the proposed DTRAD model exhibits the highest performance compared to all the other models, exhibiting the high efficiency of the proposed model.…”

Section: Results Analysismentioning

confidence: 99%

Decentralized Time-Window Based Real-Time Anomaly Detection Mechanism (DTRAD) in Iot

Sanjith¹,

Raj²

2019

IJRTE

View full text Add to dashboard Cite

Abstract: Detecting intrusions has become a mandatory service in IoT environments. This is due to the power and resource con-strained nature of the networks. This paper presents a Decentralized Time-Window based Anomaly Detection (DTRAD) model for cost and time effective intrusion detection in IoT environments. The proposed model is composed of time win-dow based training data selection module, which enables better detection and reduced bias. Training data are selected based on their temporal significance and the bag creation process is also temporally performed such that data with similar temporal signatures are grouped into same bags. The ensemble model is created and weighted voting is performed to ena-ble better results. The data reinforcement module enables new data to be appended to the training data, hence maintaining the recency of the data. Further, the entire process is decentralized, hence enabling data processing at appropriate nodes. This keeps the size of the training data low, hence reducing the computational complexity of the model to a large extent. Experiments were performed with benchmark data and comparisons were performed with recent models. Results indicate high performance of the proposed models.

show abstract

Section: Results Analysismentioning

confidence: 99%