Deep learning techniques have made early strides in the analysis about complex ECG signals, particularly in the classification about heartbeats & the detection about arrhythmias. Nonetheless, there is still more work toward be done in terms about the analysis about health-related data. This study offers dual structured & bidirectional approaches for classifying arrhythmias that deal with the drawbacks about multilayered dilated models. The data is first preprocessed using the quicker Chebyshev Type II filtering method, which does not make use about statistical properties. Using the Daubechies wavelet, which may resolve fractal issues & signal discontinuities, noise from the preprocessed filter is additionally eliminated. In this paper, the proposed models LSTM, BILSTM, & GRU were employed toward provide fusion features. The signals are categorized by fully connected layer before. The suggested model is trained & validated using the dataset for supra-ventricular arrhythmias. The learnt model considerably enhances the classification performance & interpretability by fusing dilated BILSTM with fusion features. The results about the experiment indicate that arrhythmia can be recognized using a highperformance automated recognition system. Our future development will concentrate on the automatic & cloud-based ECG classification about various arrhythmia signal-based data.