A modified LOF based approach for outlier characterization in IoT

Lynda Boukela (1) , Gongxuan Zhang (1) , Meziane Yacoub (2) , Samia Bouzefrane (3) , Sajjad Bagheri (1) , Hamed Jelodar (1)

1 NJUST - Nanjing University of Science and Technology
2 CEDRIC - MSDMA - CEDRIC. Méthodes statistiques de data-mining et apprentissage
3 CEDRIC - ROC - CEDRIC. Réseaux et Objets Connectés

The Internet of Things (IoT) is a growing paradigm that is revolutionary for Information and Communication Technology (ICT) because it gathers numerous application domains by integrating several enabling technologies. Outlier detection is a field of tremendous importance, including in IoT. In previous works on outlier detection, the proposed methods mainly tackled the efficacy and the efficiency challenges. However, a growing interest in the interpretation of the detected anomalies has been noticed by the research community, and some works have already contributed in this direction. Furthermore, characterizing anomalous events in IoT-related problems has not been conducted. Hence, in this paper, we introduce our modified Local Outlier Factor (LOF)-based outlier characterization approach and apply it to enhance the IoT security and reliability. Experiments on both synthetic and real-world datasets show the good performance of our solution.

Domaines: Réseaux et télécommunications [cs.NI]; Apprentissage [cs.LG]

Lynda Boukela, Gongxuan Zhang, Meziane Yacoub, Samia Bouzefrane, Sajjad Bagheri, et al.. A modified LOF based approach for outlier characterization in IoT. Annals of Telecommunications - annales des télécommunications, 2021, 76 (3-4), pp.145-153. ⟨10.1007/s12243-020-00780-5⟩. ⟨hal-03381644⟩