Arrhythmias refer to irregular heart rhythms, which can signal severe heart issues. With more smartwatches now able to monitor heart rates continuously, there is a new possibility to utilize this data for early detection of arrhythmias. This research introduces an ensemble learning method for classifying arrhythmias by analyzing heart rate signals collected from smartwatches. We created a multi-model ensemble that combines k-nearest neighbor, decision tree, and light gradient boosting machine models. The ensemble was trained and assessed on a dataset of 252 heart rate records from individuals over 20 years of age, which included both regular and arrhythmic cases. Our method achieved a total accuracy of 94.59%, along with a sensitivity of 89.19% and a specificity of 100%. This approach surpassed single models, such as logistic regression, support vector machines, and neural networks, illustrating the effectiveness of ensemble learning in enhancing arrhythmia detection with accessible smartwatch technology. With an F1-score of 94.29%, the ensemble model demonstrated strong capability in accurately identifying both normal and arrhythmic cases. This study advances the development of non-invasive, continuous cardiac monitoring tools that could enable early interventions and better health outcomes, potentially transforming cardiac care through widely available consumer-grade wearable devices.