Optimization of KNN Classification for ECG Data Analysis: Comparative Study of Model Performance Using the Hyperparameter Tuning and Cross-Validation

Abnormal heart rhythms such as arrhythmias can lead to severe complications, including stroke and cardiac arrest. Electrocardiography (ECG) is commonly used to monitor heart activity due to its non-invasive, affordable, and efficient nature. However, manual ECG interpretation can be time-consuming and error-prone, especially in high-demand clinical settings. This study aims to improve the performance of the K-Nearest Neighbors (KNN) algorithm for ECG signal classification by applying hyperparameter tuning and validating the results through cross-validation. ECG data were collected from participants under three physical activity conditions: sitting, walking, and running. The methodology included signal preprocessing, model development, hyperparameter tuning via Grid Search, and performance validation using K-fold cross-validation. The baseline KNN model achieved an accuracy of 78%. After optimization—by setting the number of neighbors to 16, using the Manhattan distance metric, and applying distance-based weighting—accuracy improved to 82%. Precision increased from 0.79 to 0.82, and the F1-score rose from 0.76 to 0.79. These results demonstrate the impact of systematic tuning on classification performance. An optimized KNN model offers a practical diagnostic aid for arrhythmia detection, particularly in settings with limited access to expert analysis. Its simplicity and low computational cost make it suitable for integration into portable diagnostic devices