This work proposes a method that combines machine learning and signal processing techniques to detect and classify cardiac arrhythmias in ECG signals, using the MIT-BIH Arrhythmia dataset. The methodology consists of the following steps: preprocessing to eliminate atypical morphologies in MLII derivation or accelerated heart rhythm, segmentation of the R-Peak annotations, and arrhythmias classification using 1D CNN (10 different types of arrhythmias were classified). The model achieved the following performance: accuracy and precision - 99.40%, recall - 99.32%. For explaining the results, the Grad-CAM technique was used for interpretability, identifying relevant ECG regions in the decisions. The results prove the effectiveness of the method in the automatic detection of arrhythmias.