Proposed in this work are 3 distinct advanced heart-beat arrhythmia classification algorithms that further progress recent state-of-the-art. The architectures presented achieve increases in performance by nearly all metrics and reductions in computational cost and complexity. The proposed purely neural network algorithm attained gains in both peak and average metrics, reaching scores for weighted accuracy of 98.351% and weighted F1-score of 98.362%, in addition to time reductions for epoch training and model fitting of 11.80% and 23.09%, respectively. The two compressed sensing neural network algorithms see significant gains in the metrics of weighted accuracy and F1-score as well. A comprehensive comparison study is performed against that of previous networks, quantifying the advantages of the architectural evolution.