Identifying Unusual Human Movements Using Multi-Agent and Time-Series Outlier Detection Techniques

Abstract

This research paper has introduced knowledgedriven multi-agent technology for automated machine learning in time series analysis in the context of human mobility. The main objective of this research is to identify unusual human mobility using Time Series outlier detection techniques with a more efficient multi-agent system. Detection of unusual human movement can be helpful for many domains, such as security, marketing, and health. A mobile dataset in Hiroshima, Japan between 2019-December to 2020-November was used for this research. The mobile dataset was converted to time series for multiple locations in Hiroshima, Japan. Since many different parameters are selected for time series, the message space multiagent technique is used. Sub agents are introduced for duplicate removal, missing data replacement, and outlier detection. Multiple processing agents and a control agent were introduced to predict the missing values to improve the efficiency of the model. Finally, using the Seasonal-Trend decomposition techniques, unusual movements are identified, and unusual human movements are plotted with the holidays. Multiple outlier points were detected for all the locations, and there were more than a hundred outlier points were detected for the selected locations.