In this paper we modified fuzzy-neuro generalized learning vector Quantization for Arrhyth mia heart beat detection. The original algorithm was used triangle membership function. In this research we propose another membership function is Pi membershipfunction, the Pi membership function and z membership function was adapted from twin sigmoid membership function recobnition rate of the classifier is able to be enhanced. The overall clasification system are comprised of three components including data pre-processing, feature extraction and classification system. Data processing related to how the initial data prepared, while for the feature extraction and selection, we using wavelet algorithm. From experiments show perform of a new extension can increasing the accuracy classifier compared. with the original FNGLVQ with triangle membership function. The average accuracy of original FNGLVQA comparison with FNGLVQ with Pi membership function is 94.3% and 97.96%. Also precision and recall for both algorithm respectively, 93.49% and 86.23% for oroginal FNGLVQ and 94.16% and 9.75% for FNGLVQ with Pi membership function