Parallel Support Vector Configuration for Identification of Fast Independent Components in Morphological Patterns Derived by Cardiovasographic Analysis on the Radial Pulse

The paper proposes identification of eight morphologically different pulse waveforms obtained from cardiovasographic observations on the radial pulse. Eight patterns correlated with normal subjects and those suffering from diseases of heart, lungs and liver are analyzed by genotyping of their independent components under the umbrella of parallel multiclass support vector architecture. The algorithm is computationally faster on the basis of its lower number of support vectors (10%) besides providing higher accuracy (81.6%) as compared with parallel multiclass architecture. The confusion parameters for Fast ICA based SVM suggest 3% improvement in the sensitivity and specificity causing similar improvements in the F scores and positive predictive value with a decrease in the negative likelihood ratio indicating a reduction in the number of false positives.