Ten lectures on wavelets
IEEE Transactions on Pattern Analysis and Machine Intelligence
| Hi-index | 0.00 |
Complex fractionated atrial electrograms (CFAEs) represent the electrophysiologic substrate for atrial fibrillation (AF). Individual signal complexes in CFAEs reflect electrical activity of electrophysiologic substrate at given time. To identify CFAEs sites, we developed algorithm based on wavelet transform allowing automated feature extraction from source signals. Signals were ranked by three experts into four classes. We compiled a representative data set of 113 instances with extracted features as inputs and average of expert ranking as the output. In this paper, we present results of our GAME data mining algorithm, that was used to (a) predict average ranking of experts, (b) classify into three classes. The performance of the GAME algorithm was compared to well known data mining techniques using robust ten times tenfold cross validation. Results indicate that wavelet signal decomposition could carry high level of predictive information about the state of electrophysiologic substrate and that the GAME algorithm outperforms other data mining techniques (such as decision trees, linear regression, neural networks, Support Vector Machines, etc.) in both prediction and classification accuracy.