The non-stationary impulses of an electrocardiogram (ECG) are widely utilized to assess the pace and frequency of heartbeats. An electrocardiogram (ECG) is a procedure that measures the heart's electrical activity to detect irregularities. Medical experts more widely use automatic ECG categorization in medical diagnostics and therapeutics. This work presents practical strategies for automatically classifying ECG data into two categories: standard and affected (abnormal) patients. To represent the ECG signal, morphological characteristics are retrieved for these classes. In this work, the signals are prepared with the normalization and filtering approaches, which minimizes the noise incidence. The preprocessed image is segmented, and feature vectors are selected using Ant Colony Optimization (ACO). With the assistance of feature vectors, abnormalities in the ECG signals are classified with the Bi-Long Term Short-Term approach (Bi-LSTM). The performance of the proposed approach is investigated using the performance metrics and acquired accuracy of 90%, which outperforms the existing ECG classification techniques.