A multiclass cardiac events classifier using clustering and modified adaptive neuro-fuzzy inference system

Soft computing techniques have emerged as a highly synergistic, computationally appealing, and conceptually unified framework supporting intelligent system design and analysis. The key contributing technologies of soft computing are neural network computing, fuzzy inference system, genetic algorithm or fusion of these techniques. This work proposes a method of analyzing cardiac signal (ECG) to diagnose cardiac events using adaptive neuro fuzzy inference system (ANFIS) with scaled conjugate gradient (SCG). The wavelet energy gradient algorithm is used for delineation and extraction of QRS complex from ECG signal. Seven morphological features and three statistical features are extracted from QRS complex. Optimization of the feature set is done using K-means clustering. These optimized features are used for training the ANFIS network. The experimentation is carried out for six class classifications, analysis of classifier results into average sensitivity of 95.5%, specificity of 99.13%, positive productivity of 95.52% and classification rate 98.51% The proposed methods results are compared with traditional ANFIS method with subtractive clustering, and it is found that there is a significant improvement in performance parameters of adaptive neuro-fuzzy inference system.