Implicit Biometric Authentication with Smartwatches

Today's technology enables biometric data to be obtained through wearable devices. As the popularity of wearable devices increases, the data belonging to individuals on the device also increases which requires extra security and privacy precautions. Usually, these wearable devices such as smartwatches authenticate users using PIN codes, fingerprints or Bluetooth pairing. This way of authenticating happens only one-time and it is done explicitly which may bring privacy or security concerns like shoulder-surfing, unauthorized access to authenticated devices, etc. Therefore implicit and continuous authentication is necessary. Due to mobility constraints, most of the sensors such as the face, voice, fingerprints, electroencephalography(EEG), heart signal and phonecardiography(PCG) fail to be used. In this project, we focus on recognizing and authenticating an individual with photoplethysmography sensor, electrodermal activity sensor, 3-axis accelerometer, optical thermometer. With the features we have obtained from the sensors, we have examined several Machine Learning methods, namely Logistic Regression, Random Forest, Multilayer Perceptron,Latent Dirichlet Allocation and Xgboost, for identifying an indvidual among all individuals. As a result of experiments, Xgboost algorithm gives the best results for identifying an individual. And also, we have observed that the algorithm gives better result for identifying an individual when we have data collected from much more people.

Poster: https://drive.google.com/file/d/1emqy6QxFPWyJtGT1y0B6J2kXqpxSo3C4/view?usp=sharing 

Video: https://drive.google.com/file/d/1JGNLDeBKZS0VBWdc2YUXLBGZPx66VWv6/view?usp=sharing