Information We Can Extract About a User From 'One Minute Mobile Application Usage'

TL;DR

This paper demonstrates how smartphone sensors and machine learning can identify human activities and potentially breach privacy, highlighting both the technical approach and security implications.

Contribution

It introduces a system that uses smartphone sensor data and ML algorithms for activity recognition and privacy analysis, which is a novel combination.

Findings

Sensor data can accurately classify human activities.

Feature importance analysis reveals key predictors for activity labels.

ML models can extract sensitive information, posing privacy risks.

Abstract

Understanding human behavior is an important task and has applications in many domains such as targeted advertisement, health analytics, security, and entertainment, etc. For this purpose, designing a system for activity recognition (AR) is important. However, since every human can have different behaviors, understanding and analyzing common patterns become a challenging task. Since smartphones are easily available to every human being in the modern world, using them to track the human activities becomes possible. In this paper, we extracted different human activities using accelerometer, magnetometer, and gyroscope sensors of android smartphones by building an android mobile applications. Using different social media applications, such as Facebook, Instagram, Whatsapp, and Twitter, we extracted the raw sensor values along with the attributes of $29$ subjects along with their attributes (class labels) such as age, gender, and left/right/both hands application usage. We extract features from the raw signals and use them to perform classification using different machine learning (ML) algorithms. Using statistical analysis, we show the importance of different features towards the prediction of class labels. In the end, we use the trained ML model on our data to extract unknown features from a well known activity recognition data from UCI repository, which highlights the potential of privacy breach using ML models. This security analysis could help researchers in future to take appropriate steps to preserve the privacy of human subjects.