Abstract:
In modern era IoT devices are gaining popularity in field of medical, especially when it comes to
the case of monitoring patient’s health in real time IoT devices have lot of applications in
monitoring and diagnosing. The recent cyber-attacks on IoT has arisen many security concerns
about their security. Specially, in medical field a compromised IoT device may put life of patient
at risk. Due to limitation in memory and processing resources, implementing modern security
protocols and algorithms is a big challenge. Conventional biometric technique i.e. finger print, iris
etc. are not feasible in medical field because these are one-time authentication and leave gates
open for post-authentication attacks. secondly, these biometric easily spoofed and misused. It is
safer if we authenticate patient from data of sensors monitoring its health. With this in mind ECG
based authentication is proposed, as ECG is one of the basic test used to determine patient’s
health.
In the proposed model we test and study different non-fiducial feature extraction techniques and
test them with five classifiers. For evaluation ECG-ID database consist of 89 subjects is used.
features are extracted by applying four signal transformations i.e. DCT, FFT, Time Domain and
wavelet transform. The non-fiducial approach for authentication are easy to implement but has
huge feature space and require large computational resources. To overcome this curse of
dimensionality eight feature selection techniques were tested along with five classifiers to
evaluate accuracy and execution time to find best feature subset with classifier to achieve good
accuracy with minimum execution time. The final results show that SVM is robust and less
sensitive to feature-set but require long execution time. However, Accuracy of Naïve Bayes and
Template Matching is sensitive to features but these algorithms are fast.
