شماره ركورد كنفرانس :
3540
عنوان مقاله :
Threshold-Based Hidden Markov Model for Detecting Anomalies in SAX-Represented ECG
Author/Authors :
Milad Zandi-Goharrizy Information Processing and Knowledge Discovery Laboratory (IPKD Lab) - Electrical & Computer Engineering Dept. - Yazd University, Yazd, Iran , Mohammad-Reza Zare-Mirakabad Information Processing and Knowledge Discovery Laboratory (IPKD Lab) - Electrical & Computer Engineering Dept. - Yazd University, Yazd, Iran , Fatemeh Kaveh-Yazdy Information Processing and Knowledge Discovery Laboratory (IPKD Lab) - Electrical & Computer Engineering Dept. - Yazd University, Yazd, Iran
كليدواژه :
SAX representation , ECG Data , Anomaly Detection system , Hidden Markov Models
عنوان كنفرانس :
همايش بين المللي هوش مصنوعي و پردازش سيگنال
چكيده لاتين :
In this paper, we propose a HMM based novel anomaly detection
framework, which uses SAX-represented ECGs. According to experiments, typical
HMM and SAX are not good candidates for anomaly detection, because of
low resolution of SAX. However, we contribute a threshold-based hidden Markov
model which compensates for the SAX low-resolution problem. Furthermore,
our proposed threshold reduces the dependency of the model to the distribution
of hidden state by taking into account the likelihood probability of
anomalous patterns. Results of experiments demonstrate that the typical HMM
labels samples with the accuracy of 50% and our proposed model labels same
data with the accuracy of 99%.
