Kohonen-SOM LOF Approach for Anomaly Detection

Amri Muhaimin, Kartika Fithriasari

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Water loss in distribution is a serious problem for water companies, as it can result in a loss of profit. Machine learning has a method for dealing with this, for example, anomaly detection. Anomalies in the study were detected using data from March 2017 to February 2018. The data used is water consumption, and the variables such as mean water use, maximum water use, and standard deviation of water usage were obtained. Kohonen Self-Organizing Map (SOM) and Local Outlier Factor (LOF) are used because they can create a good cluster, and LOF can detect which one of them is filled with anomalies. The Kohonen SOM algorithm produced 45 groups that were considered anomalous because their silhouette width criteria were less than the group's average silhouette width. There are 45 different groups of unexpected anomalies. The LOF generated 1229 unusual consumption events, with each incident involving 579 households or customers. The frequency calculation yields 41 customers who are suspected of being anomalous. While the baseline method from the company only captures 16 anomalies out of 41 customers. This is because the company method fails to capture unusual consumption behaviors, such as monthly consumption. The anomaly detected the customer's characteristic of having an average of more than average usage of classes and sub-zones.

Original languageEnglish
Title of host publicationProceedings - 2021 IEEE 7th Information Technology International Seminar, ITIS 2021
PublisherInstitute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)9781665408073
DOIs
Publication statusPublished - 2021
Event7th IEEE Information Technology International Seminar, ITIS 2021 - Surabaya, Indonesia
Duration: 6 Oct 20218 Oct 2021

Publication series

NameProceedings - 2021 IEEE 7th Information Technology International Seminar, ITIS 2021

Conference

Conference7th IEEE Information Technology International Seminar, ITIS 2021
Country/TerritoryIndonesia
CitySurabaya
Period6/10/218/10/21

Keywords

  • anomaly detection
  • kohonen SOM
  • local outlier factor
  • silhouette

Fingerprint

Dive into the research topics of 'Kohonen-SOM LOF Approach for Anomaly Detection'. Together they form a unique fingerprint.

Cite this