Analysis of ECG5000 Electrocardiogram Signals Using Improved DTW Algorithm

The research introduces an ECG classification framework that integrates dynamic temporal window and multi-dimensional attention mechanism, improving classification performance through adaptive feature fusion and robust temporal matching. By constructing a time-frequency domain feature system, key features such as RR interval variability and spectral energy distribution are extracted, and high-discriminative features are screened using analysis of variance. Based on the temporal perception attention model, a dynamic similarity matrix is generated by integrating signal amplitude, gradient and curvature information, and the matching window width is adaptively adjusted to address the sensitivity of traditional Dynamic Time Warping (DTW) algorithm to waveform distortion. A hybrid classification model is constructed using the joint representation of multi-dimensional features and original signals, and the path backtracking strategy is optimized by try-catch mechanism. Experiments on the ECG5000 dataset show that the proposed method achieves a classification accuracy of 94.2%, and demonstrates lower error rates than comparative algorithms under noises with different frequencies and specifications.