Improved stampede prediction model on context-awareness framework using machine learning techniques

Abstract

The determination of stampede occurrence through abnormal behaviors is an important research in context-awareness using individual activity recognition (IAR). An application such as an intelligent smartphone for crowd monitoring using inbuilt sensors is used. Meanwhile, there are few algorithms to recognize abnormal behaviors that can lead to a stampede for mitigation of crowd disasters. This study proposed an improved stampede prediction model which can facilitate abnormal detection with k-means. It can identify cluster areas among a group of people to know susceptible places that can help to predict stampede occurrence using IAR with the help of geographical positioning system (GPS) and accelerometer sensor data. To achieve this, two research questions were formulated and answered in this paper. (i) How to determine crowd of people in an area? (ii) How to know when stampede will occur in the identified area? The experimental results on the proposed model with decision tree (DT) algorithm shows an improved performance of 98.6 %, 97.7 % and 10.9 % over 94.4 %, 95 % and 18 % in the baselines for specificity, accuracy and false-negative rate (FNR) respectively thereby reducing high false negative alarm.