Identifying various arrhythmias and effectively addressing the patient condition from electrocardiogram (ECG) is a difficult task in medical field. This research proposes a cardio arrhythmia detection (CAD) framework which is a hybrid convolutional-long short term memory (HC-LSTM). The ECG signals are considered to perform pre-processing of the signals using bandpass filter to eliminate unnecessary noisy signals. The pre-processed signals are then processed for feature extraction using Residual Network-50 (ResNet-50) to obtain high frequency and spatial recognition for the complex patterns of the ECG signals due to its convolutional layers. Other features computed include statistical, frequencies, entropy, spectral, and wavelet transform features to improve the classification of signals. The hyperparameters of the network are further fine-tuned using grid search optimization (GSO) for accurate training of the HC-LSTM model. Thus, the HC-LSTM network identifies the spatial features in sequence and modeling the temporal sequence to provide an accurate model for CAD. This hybrid approach effectively addresses the complexity of ECG signal analysis, improving detection accuracy and generalizability. The final results show that the HC-LSTM model achieved an accuracy of 99.24%, precision of 97.63%, recall of 97.23%, and f-score of 98.45% when compared to the existing methods extreme gradient boosting convolutional neural network (XGB-CNN) and wavelet transform based CNN.