Cardiovascular diseases, particularly arrhythmias, are a leading cause of death worldwide, highlighting the urgent need for advanced diagnostic tools for early detection and effective treatment. This study proposes the use of a one-dimensional Convolutional Neural Network (CNN) to classify arrhythmias into five categories: Normal (N), Right Bundle Branch Block (RBBB), Left Bundle Branch Block (LBBB), Atrial Premature Beat (APB), and Premature Ventricular Contraction (PVC), using the MIT-BIH Arrhythmia Database. Unlike traditional methods that require noise removal, the proposed model processes noisy electrocardiogram (ECG) signals directly, enhancing efficiency and robustness under natural conditions that closely resemble real-world clinical applications. The data is handled without the need for conversion into images, reducing the computational burden and improving performance speed while preserving essential information in the original signal. The ECG signals were divided into 121 samples, one of the least commonly used sample sizes in recent five-category classification, contributing to reduced computational complexity without compromising classification accuracy. To address the issue of data imbalance, the SMOTE-Tomek technique was applied to improve the model’s performance, resulting in an overall accuracy increase of 98.1%. The results provide a cost-effective and efficient framework for arrhythmia classification and address the issue of data distribution imbalance, making this approach cost-effective in terms of computational cost, with high accuracy in classifying arrhythmias.