A Privacy Preserving Data Mining Methodology for Dynamically Predicting Emerging Human Threats

Abstract

This paper proposes a privacy preserving data mining driven methodology for predicting emerging human threats in a public space by capturing large scale, real time body movement data (spatial data represented in X, Y, Z coordinate space) using Red-Green-Blue (RGB) image, infrared depth and skeletal image sensing technology. Unlike traditional passive surveillance systems (e.g., CCTV video surveillance systems), multimodal surveillance technologies have the ability to capture multiple data streams in a real time dynamic manner. However, mathematical models based on machine learning principles are needed to convert the large-scale data into knowledge to serve as a decision support system for autonomously predicting emerging threats, rather than just recording and observing them as they occur. To this end, the authors of this work present a privacy preserving data mining driven methodology that captures emergent behavior of individuals in a public space and classifies them as a threat or not a threat, based on the underlying body movements through space and time. An audience in a public environment is presented as the case study for this paper with the aim of classifying individuals in the audience as threats (or not), based on their temporal body behavior profiles.

Machine learning classification of design team members' body language patterns for real time emotional state detection

“…In other works, cues such as inferring respiratory rates (by the rising and falling of the chest) and detecting fidgeting (by detecting rapid oscillations of a person's knee), have been explored (Burba, Bolas, Krum, & Suma, 2012). Manohar and Tucker utilized skeletal joint data to predict the emergence of human threats in an audience (Manohar & Tucker, 2013). Such techniques illustrate the potential of machine learning methods using skeletal data.…”

Machine learning classification of medication adherence in patients with movement disorders using non-wearable sensors

“…If the predictive accuracy of the data mining model falls below this predictive accuracy, then the new patient gait data will be included in the training data, with the model retrained using the same steps originally used during model generation. Manohar and Tucker have demonstrated the feasibility of using non-wearable sensors for modelling human body movement data in order to accurately perform emerging threat detection [34]. Behoora and Tucker have discovered correlations between engineering designers’ body language and their emotional states [35,36].…”

A data mining methodology for predicting early stage Parkinson's disease using non-invasive, high-dimensional gait sensor data

“…As new incoming patient data is acquired, the models generated by the data mining algorithms employed in this work can be iteratively updated, once their predictive accuracies fall below a given threshold (determined by the healthcare decision makers). Using non-invasive sensors, the authors have demonstrated the feasibility of modeling body movement data in this manner to predict emerging human behavior patterns (for security applications) and emotional states (for engineering education applications) with relatively high accuracy (Manohar and Tucker 2013; Behoora and Tucker 2014; Behoora and Tucker 2015). The authors have also demonstrated that these non-invasive sensors are capable of modeling and predicting medication adherence in patients with neurologically induced movement disorders (C.…”