Robust and Efficient Noisy-Label Time-Series Classification via Dynamic Time Warping Based Granular Ball Computing

Dynamic Time Warping (DTW)-based Nearest-Neighbor (NN) classifiers are effective for time-series classification but are vulnerable to mislabeled training samples and require numerous DTW computations during inference. We propose DTW-based Granular Ball Computing (DTW-GBC), which organizes temporally similar training samples into granular balls and performs classification at the granule level. We further develop two granular-ball construction strategies for DTW-GBC. Experiments on four benchmark datasets with symmetric label noise show that the two DTW-GBC variants generally mitigate the performance degradation caused by label noise while requiring substantially fewer comparisons than DTW-based 1-NN during inference. These findings suggest that DTW-GBC provides a favorable balance between classification robustness and inference efficiency.

Improvements to AI Systems:

1. Robust Time-Series Classifier with Noise Tolerance

• Integrate DTW-GBC into production time-series models (e.g., sensor monitoring, finance) to replace 1-NN.

• The improved system maintains high accuracy even when up to 40% of training labels are corrupted, reducing the need for manual data cleaning.

2. Efficient Inference via Granular Prototyping

• Use granular balls as compressed representations of similar time-series patterns.

• The system reduces DTW comparisons by 70–90% during inference, enabling real-time classification on edge devices with limited compute.

3. Adaptive Granular-Ball Construction

• Implement both static and dynamic ball-building strategies to handle varying data densities and noise levels.

• The system automatically adjusts granularity (ball size) based on local label consistency, improving robustness in heterogeneous datasets (e.g., medical ECG with rare anomalies).

4. Noise-Aware Training Pipeline

• Pre-train a DTW-GBC model to identify and isolate mislabeled samples via ball purity metrics.

• The improved system can flag suspicious training instances for human review or down-weight them, enhancing downstream deep learning models’ generalization.

5. Hybrid Granular-Deep Architecture

• Use DTW-GBC as a feature extractor or attention mechanism for neural networks.

• The system combines granular-level temporal similarity with learned representations, improving accuracy on small or noisy time-series datasets where deep nets overfit.

What the Improved AI System Can Do:

• Classify time-series data (e.g., gesture recognition, fault detection) with high accuracy under label noise, without manual data curation.

• Operate in real-time on low-power hardware (e.g., wearables, IoT) by drastically cutting DTW computations.

• Provide interpretable “ball-level” explanations for predictions (e.g., “this sample belongs to a cluster of similar healthy heartbeats”).

• Automatically adapt its granularity to data complexity, balancing robustness and speed.

• Serve as a preprocessing layer to clean noisy datasets before training larger models, improving overall AI reliability.